A versioned research record
What changed
Data additions and corrections from committed repository history.
2026-11-205e399a2 ↗
Record T-Head deployment claims and CXMT supply negotiations
- Updated: CXMT
3 field changes
- facts: [{"label":"4F2 RDIMM sampling clarification","value":"CXMT specifies sampling of its first 4F2 DDR5 RDIMM by year-end 2026; volume production depends on customer validation and demand. This is a roadmap target, not completed shipment or qualification.","evidence":"vendor","as_of":"2026-10-30","source_ids":["cxmt-rdimm-sampling-2026-10-31"]},{"label":"Sixth-generation DRAM engineering validation","value":"CXMT reports completion of engineering-sample validation for DRAM developed on its sixth-generation platform. Its unaudited nine-month results show RMB 269.024 billion revenue and RMB 132.040 billion attributable net profit, within the earlier forecast. Product specifications, customer qualification, yield and production volume remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["cxmt-q3-2026-10-31"]},{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}] → [{"label":"Long-term DRAM supply negotiations","value":"The Paper reports CXMT is negotiating long-term supply agreements with multiple leading customers. Counterparties, volumes, pricing, duration and signed-contract status are undisclosed.","evidence":"reported","as_of":"2026-11-17","source_ids":["cxmt-lta-2026-11-17"]},{"label":"4F2 RDIMM sampling clarification","value":"CXMT specifies sampling of its first 4F2 DDR5 RDIMM by year-end 2026; volume production depends on customer validation and demand. This is a roadmap target, not completed shipment or qualification.","evidence":"vendor","as_of":"2026-10-30","source_ids":["cxmt-rdimm-sampling-2026-10-31"]},{"label":"Sixth-generation DRAM engineering validation","value":"CXMT reports completion of engineering-sample validation for DRAM developed on its sixth-generation platform. Its unaudited nine-month results show RMB 269.024 billion revenue and RMB 132.040 billion attributable net profit, within the earlier forecast. Product specifications, customer qualification, yield and production volume remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["cxmt-q3-2026-10-31"]},{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}]
- source_ids: ["cxmt-rdimm-sampling-2026-10-31","cxmt-q3-2026-10-31","cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"] → ["cxmt-lta-2026-11-17","cxmt-rdimm-sampling-2026-10-31","cxmt-q3-2026-10-31","cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]
- reviewed_at: 2026-11-02 → 2026-11-20
- Updated: Alibaba T-Head
3 field changes
- facts: [{"label":"Proposed separate listing","value":"Alibaba’s November 10 filing discloses T-Head’s September 11 confidential Hong Kong listing application. Exchange clearance to proceed with the spin-off is distinct from final listing approval. T-Head is expected to remain an Alibaba subsidiary; offering size, structure and ownership dilution were not finalized. Listing Committee approval, CSRC filing and final company decisions remain conditions, with no assured completion date.","evidence":"primary","as_of":"2026-11-10","source_ids":["thead-spinoff-2026-11-10"]},{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]},{"label":"SAIL developer-access announcement","value":"Jiemian reported the July 18 WAIC announcement of SAIL developer access, including SDK, kernel-driver, profiling and debugging resources. This historical announcement is backfilled on November 9; source availability, licensing, release completeness and migration/performance claims were not independently verified.","evidence":"reported","as_of":"2026-07-18","source_ids":["thead-sail-announcement-2026-07-18"]}] → [{"label":"V900 supernode commercialization target","value":"Alibaba targets commercial-scale V900 supernode clusters in Q2 2027 and says Qwen is training its next model on T-Head chips. Configuration, time-to-quality, utilization and power are undisclosed.","evidence":"vendor","as_of":"2026-11-19","source_ids":["alibaba-compute-call-2026-11-19"]},{"label":"Proposed separate listing","value":"Alibaba’s November 10 filing discloses T-Head’s September 11 confidential Hong Kong listing application. Exchange clearance to proceed with the spin-off is distinct from final listing approval. T-Head is expected to remain an Alibaba subsidiary; offering size, structure and ownership dilution were not finalized. Listing Committee approval, CSRC filing and final company decisions remain conditions, with no assured completion date.","evidence":"primary","as_of":"2026-11-10","source_ids":["thead-spinoff-2026-11-10"]},{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]},{"label":"SAIL developer-access announcement","value":"Jiemian reported the July 18 WAIC announcement of SAIL developer access, including SDK, kernel-driver, profiling and debugging resources. This historical announcement is backfilled on November 9; source availability, licensing, release completeness and migration/performance claims were not independently verified.","evidence":"reported","as_of":"2026-07-18","source_ids":["thead-sail-announcement-2026-07-18"]}]
- source_ids: ["thead-spinoff-2026-11-10","thead-sail-announcement-2026-07-18","m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"] → ["alibaba-compute-call-2026-11-19","thead-spinoff-2026-11-10","thead-sail-announcement-2026-07-18","m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"]
- reviewed_at: 2026-11-12 → 2026-11-20
- Added: Alibaba updates domestic-compute deployment and roadmap
- Added: CXMT discusses long-term DRAM agreements
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-19 → 2026-11-20
- last_automated_run_at: 2026-11-19 → 2026-11-20
- coverage: November 19 partial sweep screened 12 priority companies and five oldest remaining profiles. Only InnoLight received a substantive update and a new profile review date. → November 20 partial sweep screened 12 priority companies, five oldest remaining profiles and T-Head. Only CXMT and T-Head received new facts and profile review dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-18","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","h3c","biren"],"summary":"English/Chinese overlap searches covered November 10–18, with older-source follow-ups. Read the RLinf/Biren co-authored account and merged SUPA PR; added public upstream support without accepting independent performance parity. Read H3C science-adoption announcement and recorded attributed university coverage. Other searches mostly repeated prior disclosures or market commentary; RSLaser commentary, SIOM irradiation news and CIOMP awards did not establish new qualified lithography deliveries. A DIGITIMES CXMT customer-validation lead was not fully verified and remains a follow-up. No new Zhiqi delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full audit. InferenceX still disclosed an unofficial deployment; no new benchmark was accepted. Coverage remains partial for private deployments, unindexed filings, tenders, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","h3c-science-adoption-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-19","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","kingsemi"],"summary":"English/Chinese overlap searches covered November 11–19. Read InnoLight November 18 primary filing and recorded the conditional buyback without interpreting it as capacity expansion. Other priority-company searches mainly repeated existing disclosures. CXMT November 17 briefing accounts and the USCC synthesis require further primary-source reconciliation; no new HBM, yield or signed-contract claim was accepted. CNFIN full-page retrieval failed. RSLaser hiring and directories, SIOM/CIOMP institutional news and Zhiqi recruitment/technical-center recognition did not establish qualified tool deliveries. A Zhiqi facility-renovation procurement lead remains unverified. Kingsemi issuer listings showed previously dated disclosures and awards. Only InnoLight profile date advanced. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; this was a head check, not a full result audit. InferenceX retained its unofficial-deployment warning; no new matched Chinese benchmark accepted. Coverage remains partial for private deployments, tenders, unindexed filings, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["innolight-buyback-proposal-2026-11-18","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-18","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","h3c","biren"],"summary":"English/Chinese overlap searches covered November 10–18, with older-source follow-ups. Read the RLinf/Biren co-authored account and merged SUPA PR; added public upstream support without accepting independent performance parity. Read H3C science-adoption announcement and recorded attributed university coverage. Other searches mostly repeated prior disclosures or market commentary; RSLaser commentary, SIOM irradiation news and CIOMP awards did not establish new qualified lithography deliveries. A DIGITIMES CXMT customer-validation lead was not fully verified and remains a follow-up. No new Zhiqi delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full audit. InferenceX still disclosed an unofficial deployment; no new benchmark was accepted. Coverage remains partial for private deployments, unindexed filings, tenders, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","h3c-science-adoption-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-19","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","kingsemi"],"summary":"English/Chinese overlap searches covered November 11–19. Read InnoLight November 18 primary filing and recorded the conditional buyback without interpreting it as capacity expansion. Other priority-company searches mainly repeated existing disclosures. CXMT November 17 briefing accounts and the USCC synthesis require further primary-source reconciliation; no new HBM, yield or signed-contract claim was accepted. CNFIN full-page retrieval failed. RSLaser hiring and directories, SIOM/CIOMP institutional news and Zhiqi recruitment/technical-center recognition did not establish qualified tool deliveries. A Zhiqi facility-renovation procurement lead remains unverified. Kingsemi issuer listings showed previously dated disclosures and awards. Only InnoLight profile date advanced. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; this was a head check, not a full result audit. InferenceX retained its unofficial-deployment warning; no new matched Chinese benchmark accepted. Coverage remains partial for private deployments, tenders, unindexed filings, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["innolight-buyback-proposal-2026-11-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-20","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","kingsemi","thead"],"summary":"English/Chinese overlap searches covered November 12–20. Read Alibaba earnings-call transcript and original The Paper reporting on CXMT; recorded attributed roadmap/adoption claims and supply negotiations, not measured parity or signed contracts. Other searches mainly repeated earlier disclosures. SIOM/CIOMP listings concerned scientific and institutional activity; no new qualified lithography delivery was established. Oldest-profile screening was search-only, not a full source audit. USCC synthesis and Zhiqi facility-procurement leads remain unresolved. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full benchmark audit. InferenceX retained its unofficial-deployment warning. No new matched Chinese benchmark accepted. Coverage remains partial for private deployments, tenders, unindexed filings, yields and component provenance. Historical reports, corrections and forecasts preserved.","source_ids":["alibaba-compute-call-2026-11-19","cxmt-lta-2026-11-17","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Alibaba September-quarter earnings call transcript
- Added: CXMT discusses long-term supply negotiations
2026-11-19214d218 ↗
Record InnoLight conditional buyback and November 19 research coverage
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]},{"label":"NPO fabrication provenance","value":"Tower says InnoLight’s first NPO device entered early production using PH18M at its California fab; multiple future products taped out on its 300mm platform. No output, yield or customer-acceptance figures were disclosed. This early device stage does not replace InnoLight’s dated H2 2027 NPO ramp expectation.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-tower-npo-2026-11-03"]},{"label":"Supply preparation for 2027 orders","value":"The November 6 filing says long-term agreements, prepayments and new suppliers helped secure scarce materials. It reiterates that major-customer orders broadly cover 2027, with 2028 demand guidance from some customers, and expects NPO/XPO demand from H2 2027. It discloses no secured component quantities, supplier identities or qualified capacity.","evidence":"vendor","as_of":"2026-11-06","source_ids":["innolight-supply-preparation-2026-11-06"]}] → [{"label":"November capital-allocation proposal","value":"Chairman proposed a RMB 2–4 billion buyback on November 18, subject to approval and execution. The accompanying operating assurances do not establish additional qualified capacity.","evidence":"primary","as_of":"2026-11-18","source_ids":["innolight-buyback-proposal-2026-11-18"]},{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]},{"label":"NPO fabrication provenance","value":"Tower says InnoLight’s first NPO device entered early production using PH18M at its California fab; multiple future products taped out on its 300mm platform. No output, yield or customer-acceptance figures were disclosed. This early device stage does not replace InnoLight’s dated H2 2027 NPO ramp expectation.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-tower-npo-2026-11-03"]},{"label":"Supply preparation for 2027 orders","value":"The November 6 filing says long-term agreements, prepayments and new suppliers helped secure scarce materials. It reiterates that major-customer orders broadly cover 2027, with 2028 demand guidance from some customers, and expects NPO/XPO demand from H2 2027. It discloses no secured component quantities, supplier identities or qualified capacity.","evidence":"vendor","as_of":"2026-11-06","source_ids":["innolight-supply-preparation-2026-11-06"]}]
- source_ids: ["innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"] → ["innolight-buyback-proposal-2026-11-18","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"]
- reviewed_at: 2026-11-09 → 2026-11-19
- Added: InnoLight chairman proposes RMB 2–4 billion buyback
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-18 → 2026-11-19
- last_automated_run_at: 2026-11-18 → 2026-11-19
- coverage: November 18 partial sweep screened 12 priority companies and five oldest remaining profiles, plus a Biren follow-up. Only Biren and H3C received substantive profile updates and new review dates. → November 19 partial sweep screened 12 priority companies and five oldest remaining profiles. Only InnoLight received a substantive update and a new profile review date.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-18","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","h3c","biren"],"summary":"English/Chinese overlap searches covered November 10–18, with older-source follow-ups. Read the RLinf/Biren co-authored account and merged SUPA PR; added public upstream support without accepting independent performance parity. Read H3C science-adoption announcement and recorded attributed university coverage. Other searches mostly repeated prior disclosures or market commentary; RSLaser commentary, SIOM irradiation news and CIOMP awards did not establish new qualified lithography deliveries. A DIGITIMES CXMT customer-validation lead was not fully verified and remains a follow-up. No new Zhiqi delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full audit. InferenceX still disclosed an unofficial deployment; no new benchmark was accepted. Coverage remains partial for private deployments, unindexed filings, tenders, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","h3c-science-adoption-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-18","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","h3c","biren"],"summary":"English/Chinese overlap searches covered November 10–18, with older-source follow-ups. Read the RLinf/Biren co-authored account and merged SUPA PR; added public upstream support without accepting independent performance parity. Read H3C science-adoption announcement and recorded attributed university coverage. Other searches mostly repeated prior disclosures or market commentary; RSLaser commentary, SIOM irradiation news and CIOMP awards did not establish new qualified lithography deliveries. A DIGITIMES CXMT customer-validation lead was not fully verified and remains a follow-up. No new Zhiqi delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full audit. InferenceX still disclosed an unofficial deployment; no new benchmark was accepted. Coverage remains partial for private deployments, unindexed filings, tenders, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","h3c-science-adoption-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-19","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","kingsemi"],"summary":"English/Chinese overlap searches covered November 11–19. Read InnoLight November 18 primary filing and recorded the conditional buyback without interpreting it as capacity expansion. Other priority-company searches mainly repeated existing disclosures. CXMT November 17 briefing accounts and the USCC synthesis require further primary-source reconciliation; no new HBM, yield or signed-contract claim was accepted. CNFIN full-page retrieval failed. RSLaser hiring and directories, SIOM/CIOMP institutional news and Zhiqi recruitment/technical-center recognition did not establish qualified tool deliveries. A Zhiqi facility-renovation procurement lead remains unverified. Kingsemi issuer listings showed previously dated disclosures and awards. Only InnoLight profile date advanced. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; this was a head check, not a full result audit. InferenceX retained its unofficial-deployment warning; no new matched Chinese benchmark accepted. Coverage remains partial for private deployments, tenders, unindexed filings, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["innolight-buyback-proposal-2026-11-18","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: InnoLight chairman proposes share repurchase
2026-11-18fa829bc ↗
Add RLinf upstream evidence and H3C education adoption review
- Updated: Biren
3 field changes
- facts: [{"label":"Public llm-d compatibility configuration","value":"The llm-d guide documents Qwen2.5-72B-Instruct-GPTQ-Int8 on eight Biren 166M GPUs, split into one prefill and one decode replica with TP=4 each. KV transfer uses NixlConnector over RDMA with CPU buffering. The project labels this community-maintained, outside nightly end-to-end testing; it is configuration evidence, not independently measured performance or production reliability.","evidence":"primary","as_of":"2026-11-17","source_ids":["llmd-biren-compatibility"]},{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"October placement completed","value":"The October 12 filing confirms completion of 130 million H shares at HK$31.08, with HK$4.0404 billion gross and approximately HK$4.0197 billion net proceeds. This closes the previously conditional October transaction; funding availability does not establish completed procurement or qualified chip production.","evidence":"primary","as_of":"2026-10-12","source_ids":["biren-placement-completion-2026-10-12"]}] → [{"label":"Upstream RLinf SUPA support","value":"RLinf PR 1581 merged on September 20, adding device discovery, visibility, BCCL communication and runtime installation. Its author reports a PI0.5 SFT end-to-end test; this review did not reproduce it or establish performance parity.","evidence":"primary","as_of":"2026-11-18","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10"]},{"label":"Public llm-d compatibility configuration","value":"The llm-d guide documents Qwen2.5-72B-Instruct-GPTQ-Int8 on eight Biren 166M GPUs, split into one prefill and one decode replica with TP=4 each. KV transfer uses NixlConnector over RDMA with CPU buffering. The project labels this community-maintained, outside nightly end-to-end testing; it is configuration evidence, not independently measured performance or production reliability.","evidence":"primary","as_of":"2026-11-17","source_ids":["llmd-biren-compatibility"]},{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"October placement completed","value":"The October 12 filing confirms completion of 130 million H shares at HK$31.08, with HK$4.0404 billion gross and approximately HK$4.0197 billion net proceeds. This closes the previously conditional October transaction; funding availability does not establish completed procurement or qualified chip production.","evidence":"primary","as_of":"2026-10-12","source_ids":["biren-placement-completion-2026-10-12"]}]
- source_ids: ["llmd-biren-compatibility","biren-ecosystem-2026-10-10","biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08","biren-placement-completion-2026-10-12"] → ["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","llmd-biren-compatibility","biren-ecosystem-2026-10-10","biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08","biren-placement-completion-2026-10-12"]
- reviewed_at: 2026-11-17 → 2026-11-18
- Updated: H3C
3 field changes
- facts: [{"label":"Open scale-up direction","value":"H3C describes UXM module support, G-Link development and NPO/CPO interfaces with 112G/224G compatibility and a reserved 448G SerDes path.","evidence":"primary","as_of":"2026-09-04","source_ids":["h3c-odx-2026-09-04"]},{"label":"Domestic-accelerator supernode","value":"H3C says UniPoD S81000 X1 supports 40 unnamed domestic accelerators in one scale-up domain, 448 GB/s peer-to-peer bandwidth, 5,760 GB of unified-address memory and up to 28 PFLOP/s FP8. Accelerator and switch silicon, dense-versus-sparse convention, power, deployment and workload results were not disclosed.","evidence":"vendor","as_of":"2026-09-29","source_ids":["h3c-unipod-s81000-x1-2026-09-29"]},{"label":"NPO interface standardization","value":"H3C reports co-authoring an OPEN NPO electrical-connector specification covering 6.4T/7.2T optical-engine interfaces and a 224G AI Rack interconnect report. It also claims multi-vendor 224G interoperability with 1.6T OSFP modules and DAC cables; no public test harness, error-rate or power results were provided.","evidence":"vendor","as_of":"2026-10-28","source_ids":["h3c-npo-standard-2026-10-28"]},{"label":"Named university deployment","value":"H3C describes a 77-node GPU cluster at Tsinghua Shenzhen International Graduate School using R5300 servers, 200G RoCE, X20000 storage and Kubernetes-based Lingxi management. Its thousand-card framing is not an exact installed-card count; GPU models, domestic origin and independently measured operating results remain undisclosed.","evidence":"vendor","as_of":"2026-10-27","source_ids":["h3c-tsinghua-sig-2026-10-27"]}] → [{"label":"Education AI-platform adoption","value":"H3C reports deployments at seven of twelve approved national AI industry-education platform universities. Hardware counts, utilization and domestic provenance remain undisclosed.","evidence":"vendor","as_of":"2026-11-18","source_ids":["h3c-science-adoption-2026-11-13"]},{"label":"Open scale-up direction","value":"H3C describes UXM module support, G-Link development and NPO/CPO interfaces with 112G/224G compatibility and a reserved 448G SerDes path.","evidence":"primary","as_of":"2026-09-04","source_ids":["h3c-odx-2026-09-04"]},{"label":"Domestic-accelerator supernode","value":"H3C says UniPoD S81000 X1 supports 40 unnamed domestic accelerators in one scale-up domain, 448 GB/s peer-to-peer bandwidth, 5,760 GB of unified-address memory and up to 28 PFLOP/s FP8. Accelerator and switch silicon, dense-versus-sparse convention, power, deployment and workload results were not disclosed.","evidence":"vendor","as_of":"2026-09-29","source_ids":["h3c-unipod-s81000-x1-2026-09-29"]},{"label":"NPO interface standardization","value":"H3C reports co-authoring an OPEN NPO electrical-connector specification covering 6.4T/7.2T optical-engine interfaces and a 224G AI Rack interconnect report. It also claims multi-vendor 224G interoperability with 1.6T OSFP modules and DAC cables; no public test harness, error-rate or power results were provided.","evidence":"vendor","as_of":"2026-10-28","source_ids":["h3c-npo-standard-2026-10-28"]},{"label":"Named university deployment","value":"H3C describes a 77-node GPU cluster at Tsinghua Shenzhen International Graduate School using R5300 servers, 200G RoCE, X20000 storage and Kubernetes-based Lingxi management. Its thousand-card framing is not an exact installed-card count; GPU models, domestic origin and independently measured operating results remain undisclosed.","evidence":"vendor","as_of":"2026-10-27","source_ids":["h3c-tsinghua-sig-2026-10-27"]}]
- source_ids: ["h3c","h3c-odx-2026-09-04","h3c-unipod-s81000-x1-2026-09-29","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27"] → ["h3c-science-adoption-2026-11-13","h3c","h3c-odx-2026-09-04","h3c-unipod-s81000-x1-2026-09-29","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27"]
- reviewed_at: 2026-10-29 → 2026-11-18
- Added: H3C reports university AI-platform adoption
- Added: RLinf and Biren document SUPA integration and embodied RL work
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-17 → 2026-11-18
- last_automated_run_at: 2026-11-17 → 2026-11-18
- coverage: November 17 partial sweep: 12 priority companies and five oldest remaining profiles screened. Biren's public inference configuration was substantively reviewed; only its profile date advanced. → November 18 partial sweep screened 12 priority companies and five oldest remaining profiles, plus a Biren follow-up. Only Biren and H3C received substantive profile updates and new review dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-18","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","h3c","biren"],"summary":"English/Chinese overlap searches covered November 10–18, with older-source follow-ups. Read the RLinf/Biren co-authored account and merged SUPA PR; added public upstream support without accepting independent performance parity. Read H3C science-adoption announcement and recorded attributed university coverage. Other searches mostly repeated prior disclosures or market commentary; RSLaser commentary, SIOM irradiation news and CIOMP awards did not establish new qualified lithography deliveries. A DIGITIMES CXMT customer-validation lead was not fully verified and remains a follow-up. No new Zhiqi delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained 6623c6c8bac460c46a8768ac261bc0889c9c7f13; unchanged heads are not a fresh full audit. InferenceX still disclosed an unofficial deployment; no new benchmark was accepted. Coverage remains partial for private deployments, unindexed filings, tenders, yields and component provenance. Historical reports and forecasts preserved.","source_ids":["rlinf-biren-supa-pr","rlinf-biren-grpo-2026-11-10","h3c-science-adoption-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: H3C describes AI for Science infrastructure adoption
- Added: Embodied GRPO with RLinf and Biren GPUs
- Added: Biren SUPA accelerator support merged into RLinf
2026-11-177aa1932 ↗
Document Biren llm-d compatibility and November 17 research coverage
- Updated: Biren
3 field changes
- facts: [{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"October placement completed","value":"The October 12 filing confirms completion of 130 million H shares at HK$31.08, with HK$4.0404 billion gross and approximately HK$4.0197 billion net proceeds. This closes the previously conditional October transaction; funding availability does not establish completed procurement or qualified chip production.","evidence":"primary","as_of":"2026-10-12","source_ids":["biren-placement-completion-2026-10-12"]}] → [{"label":"Public llm-d compatibility configuration","value":"The llm-d guide documents Qwen2.5-72B-Instruct-GPTQ-Int8 on eight Biren 166M GPUs, split into one prefill and one decode replica with TP=4 each. KV transfer uses NixlConnector over RDMA with CPU buffering. The project labels this community-maintained, outside nightly end-to-end testing; it is configuration evidence, not independently measured performance or production reliability.","evidence":"primary","as_of":"2026-11-17","source_ids":["llmd-biren-compatibility"]},{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"October placement completed","value":"The October 12 filing confirms completion of 130 million H shares at HK$31.08, with HK$4.0404 billion gross and approximately HK$4.0197 billion net proceeds. This closes the previously conditional October transaction; funding availability does not establish completed procurement or qualified chip production.","evidence":"primary","as_of":"2026-10-12","source_ids":["biren-placement-completion-2026-10-12"]}]
- source_ids: ["biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08","biren-placement-completion-2026-10-12"] → ["llmd-biren-compatibility","biren-ecosystem-2026-10-10","biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08","biren-placement-completion-2026-10-12"]
- reviewed_at: 2026-10-29 → 2026-11-17
- Added: Biren describes distributed inference adaptation and RL integrations
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-16 → 2026-11-17
- last_automated_run_at: 2026-11-16 → 2026-11-17
- coverage: November 16 daily sweep remains partial across 17 companies. The November 9–15 closed weekly briefing was appended after source rereads and gap searches; no profile dates or forecasts changed. → November 17 partial sweep: 12 priority companies and five oldest remaining profiles screened. Biren's public inference configuration was substantively reviewed; only its profile date advanced.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-17","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese overlap searches covered November 9–17 with older-source follow-ups. Read Biren's October 10 announcement and current llm-d guide; added the vendor announcement as a backfill and documented the public eight-GPU INT8 compatibility configuration with its community/non-nightly-test limitation. No measured speedup or production reliability was inferred. RLinf, slime and verl mainline claims were not exhaustively audited. Only Biren's review date advanced. Other results mainly repeated recorded memory/foundry disclosures or surfaced market commentary. RSLaser listings did not establish new scanner acceptance; SIOM/CIOMP results concerned procurement, irradiation research and institutional news. No new Zhiqi qualified delivery was verified. MLPerf Training v6.0 HEAD remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421; Inference v6.1 advanced to 6623c6c8bac460c46a8768ac261bc0889c9c7f13. Inspected its two-commit comparison; the API's first 300 file entries concerned NVIDIA/NV_AvailableVR results and ASUS metadata, including preview-to-available changes. This capped diff is not a full result audit and establishes no matched Chinese comparison. InferenceX retained its unofficial-deployment warning; no new result was accepted. Coverage remains partial for private deployments, tenders, unindexed disclosures, qualified volumes, yields and component provenance. Reports and forecasts were preserved.","source_ids":["biren-ecosystem-2026-10-10","llmd-biren-compatibility","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Biren announces llm-d adaptation and reinforcement-learning framework integration
- Added: llm-d prefill/decode guide — Biren community compatibility configuration
2026-11-16a1c6baa ↗
Publish November 9–15 weekly evidence briefing
- Added: Higher foundry activity meets equipment constraints as chip financing advances
- Updated: data/research-state.yaml
2 field changes
- coverage: November 16 partial sweep: 12 priority companies and five oldest remaining profiles screened. No new material fact or benchmark was verified; profile review dates remain unchanged. → November 16 daily sweep remains partial across 17 companies. The November 9–15 closed weekly briefing was appended after source rereads and gap searches; no profile dates or forecasts changed.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["smee","thead","smic"],"summary":"Appended weekly-2026-11-09-complete for the fully closed November 9–15 period using the cadence helper. Reread Alibaba's SEC disclosure, SMIC's HKEX quarterly filing, SMEE's CIIE reporting and SMIC's signed earnings-call report. Supplemental English/Chinese searches checked industry and manufacturing gaps; no additional qualifying event was established. Four in-period archive events were synthesized. Earlier backfills were not treated as current-week developments, and November 16 benchmark checks were labeled post-period. This was synthesis, not a fresh full-profile sweep. Research remains incomplete for private deployments, tenders, unindexed filings, qualified capacity, yields and component provenance. No profile review dates or dated forecasts changed; all earlier editions were preserved.","source_ids":["smee-ciie-2026-11-09","thead-spinoff-2026-11-10","smic-q3-2026-11-12","smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}]
2026-11-168083d32 ↗
Record November 16 partial research sweep and verification gaps
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-15 → 2026-11-16
- last_automated_run_at: 2026-11-15 → 2026-11-16
- coverage: November 15 partial sweep: 12 priority companies and five oldest remaining profiles screened; only SMIC substantively updated. Search-only coverage did not advance other profile dates. → November 16 partial sweep: 12 priority companies and five oldest remaining profiles screened. No new material fact or benchmark was verified; profile review dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-16","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped searches covered the November 8–16 overlap, with targeted follow-ups for the oldest profiles and optical-module/equipment suppliers. Read Biren's news index and CIOMP's primary homepage; the former still surfaced October compatibility announcements and the latter institutional news and awards, not new qualified production. SIOM's research index failed to open with HTTP 502; indexed November 13 results concerned irradiation research, not scanner acceptance. RSLaser searches surfaced older unverified aggregation claims; Zhiqi, Aishengna and Yuliangsheng follow-ups did not establish a new qualified delivery. No profile review date advanced. Huawei Peerium coverage repeated the existing September roadmap; InnoLight's November 6 filing was already recorded. November 12 Eoptolink coverage reiterated order confidence and an 800G delivery outlook without new quantified shipment or acceptance data. Equipment-order commentary was not promoted to qualified capacity; underlying new primary evidence remains a follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still explicitly warned of an unofficial deployment, so no new results were accepted. No new matched frontier end-to-end comparison was verified. Coverage remains partial for unindexed disclosures, private deployments, tenders, customer acceptance, yields and component provenance. Historical reports, corrections and forecasts remain intact.","source_ids":["innolight-supply-preparation-2026-11-06","mlperf-training","mlperf-inference-6-1","inferencex"]}]
2026-11-15deb26ea ↗
Record SMIC equipment delays and dated supply outlook
- Updated: SMIC
3 field changes
- facts: [{"label":"Q3 aggregate foundry operations","value":"SMIC’s November 12 quarterly filing reports 97.3% utilization, versus 93.7% in Q2; quarter-end monthly capacity of 1,116,750 and quarterly wafer sales of 3,033,683, both expressed as eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-09-30","source_ids":["smic-q3-2026-11-12"]},{"label":"Q2 aggregate foundry operations","value":"SMIC’s unaudited August 13 release reports Q2 revenue of USD 3,005.6 million, 93.7% utilization and monthly capacity of 1,096,500 eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-08-13","source_ids":["smic-q2-2026-08-13"]}] → [{"label":"Equipment lead times and supply outlook","value":"At the November 13 earnings call, co-CEO Zhao Haijun reportedly cited 16–24-month equipment delivery times and possible disruption to 2026 capital spending, while saying the year's expansion plan remained intact. He expected tight wafer supply in 2027 and AI-memory balance around mid-to-late 2028, with compute/supporting chips still constrained. These are management statements and forecasts, without tool-specific acceptance or advanced-node yield evidence.","evidence":"reported","as_of":"2026-11-13","source_ids":["smic-supply-outlook-2026-11-13"]},{"label":"Q3 aggregate foundry operations","value":"SMIC’s November 12 quarterly filing reports 97.3% utilization, versus 93.7% in Q2; quarter-end monthly capacity of 1,116,750 and quarterly wafer sales of 3,033,683, both expressed as eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-09-30","source_ids":["smic-q3-2026-11-12"]},{"label":"Q2 aggregate foundry operations","value":"SMIC’s unaudited August 13 release reports Q2 revenue of USD 3,005.6 million, 93.7% utilization and monthly capacity of 1,096,500 eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-08-13","source_ids":["smic-q2-2026-08-13"]}]
- source_ids: ["smic-q3-2026-11-12","smic-teardown","smic-q2-2026-08-13"] → ["smic-supply-outlook-2026-11-13","smic-q3-2026-11-12","smic-teardown","smic-q2-2026-08-13"]
- reviewed_at: 2026-11-13 → 2026-11-15
- Added: SMIC reports longer equipment deliveries and forecasts continued supply pressure
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-14 → 2026-11-15
- last_automated_run_at: 2026-11-14 → 2026-11-15
- coverage: November 14 partial sweep: 12 priority companies and five oldest remaining profiles screened; JCET substantively updated with an August briefing backfill. Other profile dates remain unchanged. → November 15 partial sweep: 12 priority companies and five oldest remaining profiles screened; only SMIC substantively updated. Search-only coverage did not advance other profile dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-15","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","biren"],"summary":"English/Chinese grouped overlap searches covered November 7–15, with follow-up searches for SMIC and the oldest remaining profiles. Read STAR Market Daily's signed November 13 earnings-call report via Eastmoney and added SMIC's attributed supply outlook; the republication is not independent corroboration. Reuters' original page was inaccessible; inconsistent revenue figures in syndicated snippets were not accepted. Only SMIC's review date advanced. Other searches mainly returned existing disclosures, market commentary or insufficiently supported claims. SIOM's November research concerned irradiation/FLASH therapy; CIOMP results were institutional or general optics research. No new qualified scanner-optics, light-source acceptance or Biren production evidence was verified. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields and component provenance. Historical reports, corrections and dated Observatory forecasts were preserved.","source_ids":["smic-supply-outlook-2026-11-13","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: SMIC earnings call — equipment delivery delays and AI supply outlook
2026-11-148cfae3c ↗
Backfill JCET pilot packaging and CPO validation evidence
- Updated: JCET
3 field changes
- facts: [{"label":"Advanced-packaging investment expansion","value":"At its October 27 briefing, JCET said nine-month fixed capital expenditure was nearly RMB 8.9 billion. Its 2026 investment plan rises from RMB 11.01 billion to RMB 18.2 billion, including fixed-asset investment of RMB 17.31 billion; incremental spending mainly targets advanced-packaging lines.","evidence":"reported","as_of":"2026-10-27","source_ids":["jcet-packaging-investment-2026-10-28"]},{"label":"Computing-electronics mix","value":"First-half computing-electronics revenue grew 40.4% year over year and represented 30.1% of company revenue.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Silicon-photonics validation","value":"Company reports delivering XDFOI-based silicon-photonics engine samples that passed customer validation and working with multiple customers on packaging, thermal and reliability optimization; volume and capacity were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Planned packaging-capacity financing","value":"JCET proposes raising up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion and RMB 1.0 billion for a high-density, high-capacity memory-chip system-level packaging and test upgrade. The offering and projects remain subject to approvals and completion.","evidence":"primary","as_of":"2026-09-03","source_ids":["jcet-private-placement-plan-2026-09-04"]}] → [{"label":"Glass and panel packaging pilot work","value":"In its August 31 briefing record, JCET says glass-substrate and panel-level packaging work had advanced from early R&D to pilot-line construction, with multiple customer joint-development projects. Technical challenges remained; this does not establish an operating qualified volume line.","evidence":"vendor","as_of":"2026-08-31","source_ids":["jcet-h1-briefing-2026-08-31"]},{"label":"CPO system-validation stage","value":"The same briefing reports prior CPO light-up and ongoing customer system validation with end-system customers, plus joint reliability and optical/electrical testing capabilities. It says customer products had not yet reached large-scale production or contributed actual results; future revenue depended on validation and ramp-up. Yield and shipment quantities remain undisclosed.","evidence":"vendor","as_of":"2026-08-31","source_ids":["jcet-h1-briefing-2026-08-31"]},{"label":"Advanced-packaging investment expansion","value":"At its October 27 briefing, JCET said nine-month fixed capital expenditure was nearly RMB 8.9 billion. Its 2026 investment plan rises from RMB 11.01 billion to RMB 18.2 billion, including fixed-asset investment of RMB 17.31 billion; incremental spending mainly targets advanced-packaging lines.","evidence":"reported","as_of":"2026-10-27","source_ids":["jcet-packaging-investment-2026-10-28"]},{"label":"Computing-electronics mix","value":"First-half computing-electronics revenue grew 40.4% year over year and represented 30.1% of company revenue.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Silicon-photonics validation","value":"Company reports delivering XDFOI-based silicon-photonics engine samples that passed customer validation and working with multiple customers on packaging, thermal and reliability optimization; volume and capacity were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Planned packaging-capacity financing","value":"JCET proposes raising up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion and RMB 1.0 billion for a high-density, high-capacity memory-chip system-level packaging and test upgrade. The offering and projects remain subject to approvals and completion.","evidence":"primary","as_of":"2026-09-03","source_ids":["jcet-private-placement-plan-2026-09-04"]}]
- source_ids: ["jcet","jcet-2026-h1-report","jcet-private-placement-plan-2026-09-04","jcet-packaging-investment-2026-10-28"] → ["jcet-h1-briefing-2026-08-31","jcet","jcet-2026-h1-report","jcet-private-placement-plan-2026-09-04","jcet-packaging-investment-2026-10-28"]
- reviewed_at: 2026-10-28 → 2026-11-14
- Added: JCET describes pilot-line construction and CPO validation
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-13 → 2026-11-14
- last_automated_run_at: 2026-11-13 → 2026-11-14
- coverage: November 13 partial sweep: 12 priority companies and five oldest remaining profiles screened; SMIC and Empyrean substantively updated. Other profile review dates remain unchanged. → November 14 partial sweep: 12 priority companies and five oldest remaining profiles screened; JCET substantively updated with an August briefing backfill. Other profile dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-14","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","zhiqi","jcet"],"summary":"English/Chinese overlap searches covered November 6–14 with older-source follow-ups. Read JCET’s original seven-page half-year briefing PDF through its filing-archive mirror; exact meeting date was not stated, so the news event date remains unknown. Only JCET’s review date advanced. Eoptolink’s November flood and revenue-recognition statements repeated the already-recorded October 29 disclosure; no duplicate event was added. Other searches surfaced market commentary, previously recorded announcements and unreconciled CXMT reporting. RSLaser results included a supplier profile without newly verified acceptance data; SIOM’s November 13 research concerned proton irradiation/FLASH therapy, CIOMP’s results were institutional, and Zhiqi searches returned recruitment rather than qualified scanner optics. No new lithography production claim was accepted. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX still carried its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed sources, customer acceptance, yields and provenance. Historical reports, corrections and forecasts were preserved.","source_ids":["jcet-h1-briefing-2026-08-31","eoptolink-ir-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: JCET 2026 half-year results briefing record
2026-11-13614bd14 ↗
Record SMIC Q3 operations and Empyrean acquisition proposal
- Updated: Empyrean
3 field changes
- facts: [{"label":"Customer-flow and foundry-certification claims","value":"Empyrean says products entered core design flows at leading domestic chip-design and manufacturing customers and supported mass production. It reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm; the filing does not name foundries, customers, tape-outs, covered rule decks, sign-off correlation or production volume.","evidence":"primary","as_of":"2026-06-30","source_ids":["empyrean-2026-h1-report"]},{"label":"Proposed hardware-emulation investment","value":"On October 13 the board approved contributing RMB 150 million cash and its indirectly held 7.78% S2C stake to Wuxi Yake Hongyu as part of a proposed restructuring. Cash funding is conditional on the merger; the filing says the core hardware-emulation accelerator remains in R&D, with validation and commercialization uncertain. It does not establish transaction completion or a qualified complete digital flow.","evidence":"primary","as_of":"2026-10-13","source_ids":["empyrean-yake-s2c-investment-2026-10-13"]}] → [{"label":"Proposed digital-logic EDA acquisition","value":"The October 30 board approved a proposed investment of up to RMB 220 million to acquire no more than 51% of Qijie Technology (Shenzhen). The November 2 disclosure says the aim is to address digital-logic optimization gaps and develop next-generation AI/3DIC logic and performance optimization. Consolidation is conditional on completion; board approval does not establish a completed acquisition, integrated product flow or customer-qualified sign-off capability.","evidence":"primary","as_of":"2026-10-30","source_ids":["empyrean-qijie-board-2026-11-02"]},{"label":"Customer-flow and foundry-certification claims","value":"Empyrean says products entered core design flows at leading domestic chip-design and manufacturing customers and supported mass production. It reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm; the filing does not name foundries, customers, tape-outs, covered rule decks, sign-off correlation or production volume.","evidence":"primary","as_of":"2026-06-30","source_ids":["empyrean-2026-h1-report"]},{"label":"Proposed hardware-emulation investment","value":"On October 13 the board approved contributing RMB 150 million cash and its indirectly held 7.78% S2C stake to Wuxi Yake Hongyu as part of a proposed restructuring. Cash funding is conditional on the merger; the filing says the core hardware-emulation accelerator remains in R&D, with validation and commercialization uncertain. It does not establish transaction completion or a qualified complete digital flow.","evidence":"primary","as_of":"2026-10-13","source_ids":["empyrean-yake-s2c-investment-2026-10-13"]}]
- source_ids: ["empyrean","empyrean-2026-h1-report","empyrean-yake-s2c-investment-2026-10-13"] → ["empyrean-qijie-board-2026-11-02","empyrean","empyrean-2026-h1-report","empyrean-yake-s2c-investment-2026-10-13"]
- reviewed_at: 2026-10-26 → 2026-11-13
- Updated: SMIC
3 field changes
- facts: [{"label":"Q2 aggregate foundry operations","value":"SMIC’s unaudited August 13 release reports Q2 revenue of USD 3,005.6 million, 93.7% utilization and monthly capacity of 1,096,500 eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-08-13","source_ids":["smic-q2-2026-08-13"]}] → [{"label":"Q3 aggregate foundry operations","value":"SMIC’s November 12 quarterly filing reports 97.3% utilization, versus 93.7% in Q2; quarter-end monthly capacity of 1,116,750 and quarterly wafer sales of 3,033,683, both expressed as eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-09-30","source_ids":["smic-q3-2026-11-12"]},{"label":"Q2 aggregate foundry operations","value":"SMIC’s unaudited August 13 release reports Q2 revenue of USD 3,005.6 million, 93.7% utilization and monthly capacity of 1,096,500 eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-08-13","source_ids":["smic-q2-2026-08-13"]}]
- source_ids: ["smic-teardown","smic-q2-2026-08-13"] → ["smic-q3-2026-11-12","smic-teardown","smic-q2-2026-08-13"]
- reviewed_at: 2026-11-10 → 2026-11-13
- Added: Empyrean board approves proposed Qijie control acquisition
- Added: SMIC reports 97.3% Q3 utilization and higher wafer capacity
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-12 → 2026-11-13
- last_automated_run_at: 2026-11-12 → 2026-11-13
- coverage: November 12 partial sweep: 12 priority companies, five oldest remaining profiles and T-Head screened; ACM and T-Head substantively updated. Other profile review dates remain unchanged. → November 13 partial sweep: 12 priority companies and five oldest remaining profiles screened; SMIC and Empyrean substantively updated. Other profile review dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-13","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","empyrean","zhiqi"],"summary":"English/Chinese overlap searches covered November 5–13 with earlier-source follow-ups. Read SMIC’s November 12 HKEX quarterly filing (downloaded after web retrieval timed out) and Empyrean’s November 2 board-resolution filing, resolving two pending checks. Only these two profile review dates advanced. Other priority searches largely returned previously recorded disclosures and secondary market summaries. Zhiqi searches surfaced November 10 recruitment, not scanner qualification; SIOM/CIOMP results described research or institutional activity. No new primary lithography production evidence was verified for RSLaser, Aishengna or Yuliangsheng. Eoptolink November 11–12 expansion and delivery statements surfaced in reporting and remain a primary-source follow-up. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning; no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield and component provenance. Historical reports and dated forecasts remain unchanged.","source_ids":["smic-q3-2026-11-12","empyrean-qijie-board-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Empyrean board approves proposed Qijie control acquisition
- Added: SMIC reports 97.3% Q3 utilization and higher wafer capacity
2026-11-1223b9228 ↗
Record T-Head listing disclosure and ACM shipment evidence
- Updated: ACM Research Shanghai
3 field changes
- facts: [{"label":"Parent-reported hot SPM validation","value":"US-listed parent ACM Research says a foundry validated its 170 C single-wafer hot SPM system across more than 50 process layers. It claims 70% sulfuric-acid recycling and expects roughly 400 tons saved per tool annually. Customer identity, node, tool count, wafer volume, yield and independently verified savings were not disclosed.","evidence":"vendor","as_of":"2026-10-21","source_ids":["acm-hot-spm-recycling-2026-10-21"]}] → [{"label":"Parent-level shipment measure","value":"US-listed parent ACM Research reports Q3 shipments of USD 343.5 million, up 30.6%, and revenue of USD 367.7 million. Shipments include first tools awaiting customer acceptance as well as deliveries for current-quarter revenue. These consolidated parent figures are not standalone Shanghai results or a count of accepted production tools. Customer, node, yield and tool-count breakdowns are not supplied by this aggregate measure.","evidence":"vendor","as_of":"2026-09-30","source_ids":["acm-parent-q3-2026-11-05"]},{"label":"Parent-reported hot SPM validation","value":"US-listed parent ACM Research says a foundry validated its 170 C single-wafer hot SPM system across more than 50 process layers. It claims 70% sulfuric-acid recycling and expects roughly 400 tons saved per tool annually. Customer identity, node, tool count, wafer volume, yield and independently verified savings were not disclosed.","evidence":"vendor","as_of":"2026-10-21","source_ids":["acm-hot-spm-recycling-2026-10-21"]}]
- source_ids: ["acm","acm-orders-2026-10-01","acm-hot-spm-recycling-2026-10-21"] → ["acm-parent-q3-2026-11-05","acm","acm-orders-2026-10-01","acm-hot-spm-recycling-2026-10-21"]
- reviewed_at: 2026-10-26 → 2026-11-12
- Updated: Alibaba T-Head
3 field changes
- facts: [{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]},{"label":"SAIL developer-access announcement","value":"Jiemian reported the July 18 WAIC announcement of SAIL developer access, including SDK, kernel-driver, profiling and debugging resources. This historical announcement is backfilled on November 9; source availability, licensing, release completeness and migration/performance claims were not independently verified.","evidence":"reported","as_of":"2026-07-18","source_ids":["thead-sail-announcement-2026-07-18"]}] → [{"label":"Proposed separate listing","value":"Alibaba’s November 10 filing discloses T-Head’s September 11 confidential Hong Kong listing application. Exchange clearance to proceed with the spin-off is distinct from final listing approval. T-Head is expected to remain an Alibaba subsidiary; offering size, structure and ownership dilution were not finalized. Listing Committee approval, CSRC filing and final company decisions remain conditions, with no assured completion date.","evidence":"primary","as_of":"2026-11-10","source_ids":["thead-spinoff-2026-11-10"]},{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]},{"label":"SAIL developer-access announcement","value":"Jiemian reported the July 18 WAIC announcement of SAIL developer access, including SDK, kernel-driver, profiling and debugging resources. This historical announcement is backfilled on November 9; source availability, licensing, release completeness and migration/performance claims were not independently verified.","evidence":"reported","as_of":"2026-07-18","source_ids":["thead-sail-announcement-2026-07-18"]}]
- source_ids: ["thead-sail-announcement-2026-07-18","m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"] → ["thead-spinoff-2026-11-10","thead-sail-announcement-2026-07-18","m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"]
- reviewed_at: 2026-11-09 → 2026-11-12
- Added: ACM parent reports Q3 shipments including tools awaiting acceptance
- Added: Alibaba discloses confidential T-Head listing application
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-11 → 2026-11-12
- last_automated_run_at: 2026-11-10 → 2026-11-12
- coverage: November 11 partial sweep: 12 priority companies and four oldest remaining profiles screened; Enflame and Guowang substantively updated. Other profile review dates remain unchanged. → November 12 partial sweep: 12 priority companies, five oldest remaining profiles and T-Head screened; ACM and T-Head substantively updated. Other profile review dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","ciomp","acm","empyrean","thead"],"summary":"English/Chinese overlap searches covered November 4–12, with earlier-source follow-ups. Read Alibaba’s November 10 SEC exhibit and ACM’s November 5 company release via GlobeNewswire after its IR page returned 403. Only T-Head and ACM profile review dates advanced. Other searches largely returned market-price commentary and existing disclosures. RSLaser/SIOM production claims surfaced in secondary summaries without primary verification and were not added; CIOMP indexed results described institutional activity, not scanner acceptance. Empyrean’s proposed Qijie acquisition surfaced in November 2 reporting and remains a primary-filing follow-up. CXMT reporting and meeting-notice dates require reconciliation before adding further financial or customer-agreement claims. SMIC’s indexed announcements did not yet supply the scheduled November 12 quarterly results at this morning’s check. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, unindexed sources, tenders, acceptance, yield and component provenance. Historical reports, forecasts and corrections remain unchanged.","source_ids":["thead-spinoff-2026-11-10","acm-parent-q3-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: ACM parent reports Q3 shipments including tools awaiting acceptance
- Added: Alibaba discloses confidential T-Head listing application
2026-11-1162b8f99 ↗
research: add Enflame prepayments and Guowang facility evidence
- Updated: Enflame
3 field changes
- facts: [{"label":"Long-context inference software claim","value":"In its October 29 conference release, Enflame describes operator fusion, memory optimization, scheduling, hierarchical KV Cache and prefill/decode separation, claiming stable high-throughput inference at million-token context lengths. It supplies no model, precision, accelerator count, latency, throughput, power or reproducible test configuration. This is a vendor capability claim, not a verified operating benchmark.","evidence":"vendor","as_of":"2026-10-29","source_ids":["enflame-tencent-inference-2026-10-29"]},{"label":"System partner and corrected model name","value":"Enflame's October 16 release, republished October 18, identifies the model as ESL64-O (letter O) and names ZTE as co-developer. Its claim of supporting clusters above 10,000 cards describes design capability, not a measured or accepted deployment. See the appended correction to the earlier exhibition record.","evidence":"vendor","as_of":"2026-10-16","source_ids":["enflame-ciif-release-2026-10-16"]},{"label":"ESL64-O supernode","value":"Enflame displayed an ESL64-O supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}] → [{"label":"Prepayments and operating cash needs","value":"The unaudited September 30 balance sheet reports RMB 7.709 billion of prepayments, versus RMB 388.132 million at year-end, and RMB 6.385 billion of contract liabilities, versus RMB 346.042 million. Nine-month operating cash outflow was RMB 2.017 billion, attributed to increased raw-material supplier payments. Q3 revenue was RMB 1.212 billion and the attributable net loss RMB 121.639 million. These financial balances do not identify suppliers, reserved wafer or HBM quantities, chip deliveries or a replenishable domestic supply route.","evidence":"vendor","as_of":"2026-09-30","source_ids":["enflame-q3-2026-10-31"]},{"label":"Long-context inference software claim","value":"In its October 29 conference release, Enflame describes operator fusion, memory optimization, scheduling, hierarchical KV Cache and prefill/decode separation, claiming stable high-throughput inference at million-token context lengths. It supplies no model, precision, accelerator count, latency, throughput, power or reproducible test configuration. This is a vendor capability claim, not a verified operating benchmark.","evidence":"vendor","as_of":"2026-10-29","source_ids":["enflame-tencent-inference-2026-10-29"]},{"label":"System partner and corrected model name","value":"Enflame's October 16 release, republished October 18, identifies the model as ESL64-O (letter O) and names ZTE as co-developer. Its claim of supporting clusters above 10,000 cards describes design capability, not a measured or accepted deployment. See the appended correction to the earlier exhibition record.","evidence":"vendor","as_of":"2026-10-16","source_ids":["enflame-ciif-release-2026-10-16"]},{"label":"ESL64-O supernode","value":"Enflame displayed an ESL64-O supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}]
- source_ids: ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12","enflame-ciif-release-2026-10-16","enflame-tencent-inference-2026-10-29"] → ["enflame-q3-2026-10-31","enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12","enflame-ciif-release-2026-10-16","enflame-tencent-inference-2026-10-29"]
- reviewed_at: 2026-11-03 → 2026-11-11
- Updated: Beijing Guowang
3 field changes
- facts: [{"label":"Optical manufacturing recruitment","value":"An October 16 Hebei University notice recruits 2027 graduates for optical coating, processing and inspection, optical-mechanical integration, and system testing. It supports continuing staffing activity but identifies no qualified projection-optics product, delivered unit, customer acceptance or production yield.","evidence":"primary","as_of":"2026-10-16","source_ids":["guowang-recruitment-2026-10-16"]}] → [{"label":"Yizhuang facility renovation procurement","value":"A November 2 government notice discloses qualified applicants for Building 3 renovation. The August 18 tender specifies 11,610.69 square metres, an estimated RMB 76.9104 million contract and a planned 176-day construction period. These are procurement-stage premises works, not evidence of a contract award, completed facility, qualified projection optics or production throughput.","evidence":"primary","as_of":"2026-11-02","source_ids":["guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02"]},{"label":"Optical manufacturing recruitment","value":"An October 16 Hebei University notice recruits 2027 graduates for optical coating, processing and inspection, optical-mechanical integration, and system testing. It supports continuing staffing activity but identifies no qualified projection-optics product, delivered unit, customer acceptance or production yield.","evidence":"primary","as_of":"2026-10-16","source_ids":["guowang-recruitment-2026-10-16"]}]
- source_ids: ["guowang","guowang-recruitment-2026-10-16"] → ["guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","guowang","guowang-recruitment-2026-10-16"]
- reviewed_at: 2026-10-23 → 2026-11-11
- Added: Enflame filing documents larger prepayments and cash needs
- Added: Guowang renovation tender advances to applicant disclosure
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-11-10 → 2026-11-11
- coverage: November 10 partial sweep: 12 priority companies and five oldest remaining profiles screened; SMIC and SMEE substantively updated. Other profile review dates remain unchanged. → November 11 partial sweep: 12 priority companies and four oldest remaining profiles screened; Enflame and Guowang substantively updated. Other profile review dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-11","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","ciomp"],"summary":"English/Chinese overlap searches covered November 3–11, with older-source follow-ups. Read Enflame’s original unaudited Q3 filing, resolving the prior prepayment follow-up without inferring HBM provenance or delivery quantities. Read Guowang’s November 2 qualified-applicant notice, its attachment and the August 18 tender; construction scope and estimated spending remain distinct from completed premises or qualified optics. Only Enflame and Guowang profile review dates advanced. RSLaser’s website could not be retrieved; indexed claims of volume delivery lacked a verified primary source. SIOM Chinese retrieval timed out and its English research index returned 502; indexed space-station cooling research did not establish lithography progress. Read CIOMP’s research index and September 17 telescope-mirror review; no new scanner-qualification evidence was verified. Other priority searches mainly returned already-recorded disclosures, older reporting and broker commentary. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["enflame-q3-2026-10-31","guowang-renovation-tender-2026-08-18","guowang-renovation-applicants-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Enflame 2026 third-quarter report (unaudited)
- Added: Guowang Yizhuang Building 3 renovation qualified-applicant disclosure
- Added: Guowang Yizhuang Building 3 renovation tender prequalification notice
2026-11-10b3fee82 ↗
research: add SMEE dry-tool claims and SMIC operating evidence
- Updated: SMEE
3 field changes
- facts: [] → [{"label":"Dry-lithography product scope","value":"The undated official listings specify SSB800 I-line for 280 nm and above and SSC800 KrF for 110 nm and above. These product scopes do not establish advanced immersion or EUV qualification.","evidence":"vendor","as_of":"2026-11-10","source_ids":["smee-ssb800","smee-ssc800"]},{"label":"Attributed dry-tool operating claims","value":"November 9 CIIE reporting quotes SMEE claiming over 90% domestic content in key components, more than 10 million cumulative wafer passes supported, and effective operating rate above 92%. The exhibit comprised platform models. Customer identities, installed-tool count, measurement period, uptime methodology and component-content denominator are undisclosed; repeated wafer processing must not be read as unique good wafers or independent performance validation.","evidence":"vendor","as_of":"2026-11-09","source_ids":["smee-ciie-2026-11-09"]}]
- source_ids: ["smee","duv-reuters"] → ["smee","duv-reuters","smee-ssb800","smee-ssc800","smee-ciie-2026-11-09"]
- reviewed_at: 2026-10-25 → 2026-11-10
- Updated: SMIC
3 field changes
- facts: [] → [{"label":"Q2 aggregate foundry operations","value":"SMIC’s unaudited August 13 release reports Q2 revenue of USD 3,005.6 million, 93.7% utilization and monthly capacity of 1,096,500 eight-inch-equivalent standard-logic wafers. These aggregate figures do not disclose advanced-node capacity, good large AI dies, process yield or domestic-tool share.","evidence":"vendor","as_of":"2026-08-13","source_ids":["smic-q2-2026-08-13"]}]
- source_ids: ["smic-teardown"] → ["smic-teardown","smic-q2-2026-08-13"]
- reviewed_at: 2026-09-11 → 2026-11-10
- Added: SMEE exhibits dry-lithography models and reports operating metrics
- Added: SMIC Q2 filing adds aggregate capacity and utilization evidence
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-09 → 2026-11-10
- last_automated_run_at: 2026-11-08 → 2026-11-10
- coverage: November 9 weekly briefing: eight public November 2–8 disclosures synthesized across five companies. Coverage remains partial; no additional profile review dates advanced. → November 10 partial sweep: 12 priority companies and five oldest remaining profiles screened; SMIC and SMEE substantively updated. Other profile review dates remain unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]},{"date":"2026-11-10","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","siom","smee","ciomp"],"summary":"English/Chinese overlap searches covered November 2–10, with older-source follow-ups. Read SMIC’s original unaudited Q2 filing through Sina and resolved the prior capacity backfill gap. Its October 31 notice schedules Q3 results for November 12; no Q3 actuals were inferred. Read SMEE official SSB800/SSC800 product listings and November 9 Guancha reporting through Sina, retaining vendor attribution for operating and domestic-content claims and distinguishing exhibition models from qualified advanced tools. Only SMIC and SMEE review dates advanced. RSLaser/Guowang searches returned undated or historical institutional and recruitment descriptions; no new accepted-tool evidence verified. SIOM homepage retrieval timed out; indexed institutional results did not establish a new qualified lithography source. CIOMP results included a 2016 project test, not a current milestone. Other priority searches mainly returned recorded disclosures and market commentary; broker claims about improved HBM supply and Enflame supply prepayments remain primary-filing follow-ups. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; individual result artifacts were not exhaustively audited. InferenceX retained its unofficial-deployment warning, so no new results were accepted. No new matched frontier end-to-end benchmark was verified. Coverage remains partial for private deployments, tenders, unindexed filings, acceptance, yield, reliability and component provenance. Historical reports, forecasts, corrections and benchmark records are preserved.","source_ids":["smic-q2-2026-08-13","smee-ciie-2026-11-09","smee-ssb800","smee-ssc800","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: SMEE dry-lithography models and attributed operating claims at CIIE
- Added: SSB800 product page
- Added: SSC800 product page
- Added: Unaudited results for the three months ended June 30, 2026
2026-11-09fe4c0f4 ↗
Add November 2–8 weekly briefing with scoped production evidence
- Added: Early optical production and equipment progress leave supply dependencies unresolved
- Updated: data/research-state.yaml
2 field changes
- coverage: November 9 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; AMEC, InnoLight and T-Head substantively updated. Other profile dates unchanged. → November 9 weekly briefing: eight public November 2–8 disclosures synthesized across five companies. Coverage remains partial; no additional profile review dates advanced.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["amec","hwatsing","iluvatar","innolight","ymtc"],"summary":"Prepared weekly-2026-11-02-complete using the closed-period helper for November 2–8. Synthesized eight archived in-period disclosures and performed supplemental English/Chinese industry and accelerator gap searches. Read Tower’s original NPO announcement, InnoLight’s November 3 and November 6 investor documents, Hwatsing management-call reporting and Iluvatar shipment reporting; reused today’s original AMEC document review. Reuters direct-page retrieval failed, so indexed Reuters reporting and accessible Reuters relays supported the YMTC sourcing and IPO sections, without treating relays as independent corroboration. Original Iluvatar exchange-filing retrieval and additional performance-claim verification remain unresolved. November 9 MLPerf head checks from the separately logged daily sweep and the InferenceX unofficial-deployment warning are labeled post-period; no exhaustive artifact audit or period-end leaderboard reconstruction was performed. No new benchmark, forecast revision or company review-date change was made. Preserved all historical reports and prior run records. This briefing is complete as a calendar period but research coverage remains partial, including private deployments, tenders, unindexed disclosures and qualified-production metrics. Source synchronization and publication are checked separately after commit.","source_ids":["amec-ir-2026-11-02","hwatsing-acceptance-2026-11-05","iluvatar-shipments-2026-11-02","inferencex","innolight-ir-2026-11-03","innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","mlperf-inference-6-1","mlperf-training","ymtc-domestic-tools-2026-11-06","ymtc-ipo-delay-2026-11-04"]}]
2026-11-0962e0f34 ↗
research: distinguish AMEC tool stages and record November 9 coverage
- Updated: AMEC
3 field changes
- facts: [{"label":"Reported installed reaction-chamber scale","value":"China Star Market reported that AMEC had more than 8,800 cumulative reaction chambers operating in over 220 domestic and overseas production lines at more than 100 customers by June 2026. The report did not name customers or disclose product mix, nodes, utilization, uptime, yield contribution or repeat-order economics.","evidence":"reported","as_of":"2026-06-30","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Advanced-memory tool launches","value":"AMEC displayed its newly launched Primo XD-RIE 70-to-90:1 aspect-ratio etch system and Prefima Unicore AM 300 mm molybdenum ALD system. Announced capabilities were not accompanied by named-customer qualification, accepted deliveries, production use or operating metrics.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]},{"label":"September production footprint","value":"The unaudited quarterly filing reports more than 9,300 cumulative reaction chambers in volume-production lines domestically and overseas by September 30. Nine-month revenue was RMB 10.326 billion; attributable profit excluding non-recurring items was RMB 1.623 billion. Product mix, named customers, nodes, utilization and yield contribution remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["amec-q3-2026-10-31"]}] → [{"label":"Reported installed reaction-chamber scale","value":"China Star Market reported that AMEC had more than 8,800 cumulative reaction chambers operating in over 220 domestic and overseas production lines at more than 100 customers by June 2026. The report did not name customers or disclose product mix, nodes, utilization, uptime, yield contribution or repeat-order economics.","evidence":"reported","as_of":"2026-06-30","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Advanced-memory tool launches","value":"AMEC displayed its newly launched Primo XD-RIE 70-to-90:1 aspect-ratio etch system and Prefima Unicore AM 300 mm molybdenum ALD system. Announced capabilities were not accompanied by named-customer qualification, accepted deliveries, production use or operating metrics.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]},{"label":"September production footprint","value":"The unaudited quarterly filing reports more than 9,300 cumulative reaction chambers in volume-production lines domestically and overseas by September 30. Nine-month revenue was RMB 10.326 billion; attributable profit excluding non-recurring items was RMB 1.623 billion. Product mix, named customers, nodes, utilization and yield contribution remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["amec-q3-2026-10-31"]},{"label":"Etch operating-stage distinction","value":"Company reports its 60:1 etcher in volume production with uptime above 90%; the measurement period, fleet and method are undisclosed. Its 90:1 successor remains in customer validation. Primo Angnova reportedly achieved 140:1 etching and shipped for certification, not demonstrated qualified volume production.","evidence":"vendor","as_of":"2026-11-02","source_ids":["amec-ir-2026-11-02"]},{"label":"Tungsten deposition shipments","value":"Company reports more than 500 cumulative reaction stations shipped across CVD-W, HAR-W and ALD-W products. This is not a count of complete tools or independently accepted installations.","evidence":"vendor","as_of":"2026-11-02","source_ids":["amec-ir-2026-11-02"]},{"label":"Memory demand and parts supply","value":"Company says NAND and DRAM represented about 60% of nine-month orders, while component shortages challenge delivery. Domestic-supplier development and capacity expansion are mitigation efforts, not proof of a fully domestic bill of materials.","evidence":"vendor","as_of":"2026-11-02","source_ids":["amec-ir-2026-11-02"]}]
- source_ids: ["amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13","guoxin-expo-2026-10-12","amec-q3-2026-10-31"] → ["amec-ir-2026-11-02","amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13","guoxin-expo-2026-10-12","amec-q3-2026-10-31"]
- reviewed_at: 2026-10-31 → 2026-11-09
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]},{"label":"NPO fabrication provenance","value":"Tower says InnoLight’s first NPO device entered early production using PH18M at its California fab; multiple future products taped out on its 300mm platform. No output, yield or customer-acceptance figures were disclosed. This early device stage does not replace InnoLight’s dated H2 2027 NPO ramp expectation.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-tower-npo-2026-11-03"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]},{"label":"NPO fabrication provenance","value":"Tower says InnoLight’s first NPO device entered early production using PH18M at its California fab; multiple future products taped out on its 300mm platform. No output, yield or customer-acceptance figures were disclosed. This early device stage does not replace InnoLight’s dated H2 2027 NPO ramp expectation.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-tower-npo-2026-11-03"]},{"label":"Supply preparation for 2027 orders","value":"The November 6 filing says long-term agreements, prepayments and new suppliers helped secure scarce materials. It reiterates that major-customer orders broadly cover 2027, with 2028 demand guidance from some customers, and expects NPO/XPO demand from H2 2027. It discloses no secured component quantities, supplier identities or qualified capacity.","evidence":"vendor","as_of":"2026-11-06","source_ids":["innolight-supply-preparation-2026-11-06"]}]
- source_ids: ["innolight-tower-npo-2026-11-03","innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"] → ["innolight-supply-preparation-2026-11-06","innolight-tower-npo-2026-11-03","innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"]
- reviewed_at: 2026-11-07 → 2026-11-09
- Updated: Alibaba T-Head
3 field changes
- facts: [{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]}] → [{"label":"Reported China market share","value":"IDC data reported by Reuters put T-Head at 7% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"M890 rack system","value":"Alibaba said the Panjiu AL128 packages 128 M890 accelerators in one rack and was available to Chinese enterprise customers through Bailian; operating and customer-acceptance data were not disclosed.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"Accelerator roadmap","value":"Alibaba targets Zhenwu V900 for Q3 2027 and J900 for Q3 2028; these are expected milestones, not delivered products, and the reported performance multipliers lack a disclosed workload.","evidence":"reported","as_of":"2026-05-20","source_ids":["m890"]},{"label":"V900 launch update","value":"Alibaba unveiled V900 and now targets mass production and commercial release in Q1 2027, earlier than its May Q3 2027 roadmap; it claims three times M890 performance and clusters of up to 500,000 chips without disclosing workload, precision, topology, power or operating results.","evidence":"reported","as_of":"2026-09-22","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22"]},{"label":"Supernode commercialization and supply","value":"Alibaba says M890 AI Supernodes began reaching commercial scale in the third quarter of 2026, while medium- and long-term demand exceeds its supply capability and wider data-center supply-chain shortages constrain expansion. It did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results.","evidence":"vendor","as_of":"2026-09-22","source_ids":["alibaba-v900-primary-2026-09-22"]},{"label":"SAIL developer-access announcement","value":"Jiemian reported the July 18 WAIC announcement of SAIL developer access, including SDK, kernel-driver, profiling and debugging resources. This historical announcement is backfilled on November 9; source availability, licensing, release completeness and migration/performance claims were not independently verified.","evidence":"reported","as_of":"2026-07-18","source_ids":["thead-sail-announcement-2026-07-18"]}]
- source_ids: ["m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"] → ["thead-sail-announcement-2026-07-18","m890","thead","alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","china-ai-accelerator-share-reuters-2026-10-13"]
- reviewed_at: 2026-10-13 → 2026-11-09
- Added: AMEC separates production etchers from newer validation-stage tools
- Added: InnoLight describes material commitments behind its 2027 order outlook
- Added: T-Head announced SAIL developer access at WAIC
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-11-08 → 2026-11-09
- coverage: November 8 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; only YMTC substantively updated. Other profile dates unchanged. → November 9 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; AMEC, InnoLight and T-Head substantively updated. Other profile dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-09","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches covered November 1–9, overlapping the last November 8 run by seven days, with older-source follow-ups. Read AMEC’s original October investor-record DOCX and InnoLight’s November 6 filing PDF; added attributed production-stage, shipment and supply-preparation evidence. Backfilled T-Head’s July 18 SAIL announcement from original Jiemian reporting; the Alibaba Cloud community account of phased opening was read but not treated as an audited release. Only these three profile dates advanced. T-Head developer portal returned no readable release details; source/license/release audit and the November 5 Kimi adaptation lead remain open. RSLaser searches returned undated descriptions and unsupported performance claims; no accepted-scanner qualification was established. Guowang searches returned historical or unrelated material, without verified projection-optics delivery. SIOM’s current institutional index was readable but supplied no new relevant qualified-source evidence. SMIC financial-page retrieval was blocked; Q2 capacity/earnings backfill remains unverified in primary documents. Other priority searches mainly returned previously recorded evidence or generic operating assurances. MLPerf Training v6.0 and Inference v6.1 HEADs remained eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; no exhaustive submission-artifact audit was performed. InferenceX again displayed an unofficial-deployment warning, so no new results were accepted; original-run provenance remains unresolved. No new matched frontier AI training/inference comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, acceptance, energy, reliability and component provenance. Benchmarks, historical reports and dated forecasts were preserved.","source_ids":["amec-ir-2026-11-02","innolight-supply-preparation-2026-11-06","thead-sail-announcement-2026-07-18","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: 2026年10月投资者关系活动记录表
- Added: 关于股份回购事项的公告(含经营情况及未来展望)
- Added: 平头哥开源AI软件栈T-Head SAIL
2026-11-08e18237f ↗
research: record reported YMTC financing delay and November 8 coverage
- Updated: YMTC
3 field changes
- facts: [{"label":"Capacity expansion","value":"Reuters reported two additional factories planned beyond one due for completion in 2026; completed capacity, qualified output and domestic equipment provenance remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Equipment diversification","value":"The Wire China reported that YMTC replaced blocked U.S. chipmaking tools with Dutch and Japanese equipment; suppliers, models, domestic share and complete-flow provenance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Licensed process IP","value":"The Wire China reported that at least one YMTC innovation was licensed to Samsung; the patent, terms, production use and economic contribution were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Pricing and product mix","value":"The Wire China reportedly quoted one unnamed employee saying YMTC had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. Affected products, price changes, capacity allocation and corporate guidance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06"]},{"label":"Global shipment scale","value":"Counterpoint estimated YMTC held 14% of global NAND shipment volume in the second quarter of 2026, ranking third, but ranked fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low enterprise-SSD share. A later Counterpoint article described a 16.5% current shipment share without publicly reconciling the measurement basis or revision with the earlier 14% estimate.","evidence":"measured","as_of":"2026-06-30","source_ids":["counterpoint-ymtc-q2-shipments-2026-08-12","counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"DRAM research and testing","value":"Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month and explicitly said the device had no mass-production plan. A faster 9,600 MT/s version remained a development target for Fab 3.","evidence":"reported","as_of":"2026-10-13","source_ids":["counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"Proposed STAR Market financing","value":"The Shanghai Stock Exchange accepted YMTC's application for a proposed RMB 33 billion STAR Market raise; the project was under inquiry and did not establish completed financing or installed capacity.","evidence":"primary","as_of":"2026-08-21","source_ids":["ymtc-star-market-ipo-2026-08-21"]},{"label":"Fourth-fab equipment target","value":"Reuters reports a target of at least 60% domestic equipment, with an ambition of 70%, for the planned fourth fab. Supplier talks remain preliminary, with no formal orders.","evidence":"reported","as_of":"2026-11-06","source_ids":["ymtc-domestic-tools-2026-11-06"]}] → [{"label":"Capacity expansion","value":"Reuters reported two additional factories planned beyond one due for completion in 2026; completed capacity, qualified output and domestic equipment provenance remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Equipment diversification","value":"The Wire China reported that YMTC replaced blocked U.S. chipmaking tools with Dutch and Japanese equipment; suppliers, models, domestic share and complete-flow provenance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Licensed process IP","value":"The Wire China reported that at least one YMTC innovation was licensed to Samsung; the patent, terms, production use and economic contribution were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Pricing and product mix","value":"The Wire China reportedly quoted one unnamed employee saying YMTC had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. Affected products, price changes, capacity allocation and corporate guidance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06"]},{"label":"Global shipment scale","value":"Counterpoint estimated YMTC held 14% of global NAND shipment volume in the second quarter of 2026, ranking third, but ranked fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low enterprise-SSD share. A later Counterpoint article described a 16.5% current shipment share without publicly reconciling the measurement basis or revision with the earlier 14% estimate.","evidence":"measured","as_of":"2026-06-30","source_ids":["counterpoint-ymtc-q2-shipments-2026-08-12","counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"DRAM research and testing","value":"Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month and explicitly said the device had no mass-production plan. A faster 9,600 MT/s version remained a development target for Fab 3.","evidence":"reported","as_of":"2026-10-13","source_ids":["counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"Proposed STAR Market financing","value":"The Shanghai Stock Exchange accepted YMTC's application for a proposed RMB 33 billion STAR Market raise; the project was under inquiry and did not establish completed financing or installed capacity.","evidence":"primary","as_of":"2026-08-21","source_ids":["ymtc-star-market-ipo-2026-08-21"]},{"label":"Fourth-fab equipment target","value":"Reuters reports a target of at least 60% domestic equipment, with an ambition of 70%, for the planned fourth fab. Supplier talks remain preliminary, with no formal orders.","evidence":"reported","as_of":"2026-11-06","source_ids":["ymtc-domestic-tools-2026-11-06"]},{"label":"Reported financing delay","value":"Reuters reports the IPO process is on hold. Sources suggest H1 2027, conditional on markets; this is not a confirmed listing date.","evidence":"reported","as_of":"2026-11-04","source_ids":["ymtc-ipo-delay-2026-11-04"]}]
- source_ids: ["ymtc","china-memory-expansion-reuters-2026-07-24","ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","counterpoint-ymtc-q2-shipments-2026-08-12","ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-ipo-dram-2026-10-13","ymtc-domestic-tools-2026-11-06"] → ["ymtc-ipo-delay-2026-11-04","ymtc","china-memory-expansion-reuters-2026-07-24","ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","counterpoint-ymtc-q2-shipments-2026-08-12","ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-ipo-dram-2026-10-13","ymtc-domestic-tools-2026-11-06"]
- reviewed_at: 2026-11-06 → 2026-11-08
- Added: YMTC listing reportedly delayed
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-07 → 2026-11-08
- last_automated_run_at: 2026-11-06 → 2026-11-08
- coverage: November 7 partial sweep: 12 priority-1 companies and five oldest remaining profiles screened; InnoLight, NAURA, Ligent and RoboTechnik substantively updated. Other review dates unchanged. → November 8 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; only YMTC substantively updated. Other profile dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-08","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","guowang","siom"],"summary":"English/Chinese searches overlapped October 31–November 8 and included older-source follow-ups. Read original Reuters reporting and added a financing update for YMTC; only that profile date advanced. Other priority searches chiefly returned previously recorded material. RSLaser institutional description lacked dated qualification or output evidence; Guowang results concerned historical construction, not current accepted optics. SIOM research-page retrieval timed out; search results and its contact-lithography procurement lead were not treated as advanced production evidence. T-Head developer page had no readable release details; SAIL code, license and release inspection remains unresolved. MLPerf Training v6.0 and Inference v6.1 commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No exhaustive artifact audit was performed. InferenceX again displayed an unofficial-deployment warning; no new results were accepted and original-run provenance remains a follow-up. No new matched frontier end-to-end comparison verified. Coverage remains partial for private deployments, tenders, unindexed disclosures, yields, customer acceptance, energy, reliability and component provenance. Existing benchmarks, dated forecasts and historical reports were preserved.","source_ids":["ymtc-ipo-delay-2026-11-04","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: YMTC listing reportedly delayed
2026-11-073204908 ↗
research: document photonics fabrication, capacity and equipment evidence
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]},{"label":"NPO fabrication provenance","value":"Tower says InnoLight’s first NPO device entered early production using PH18M at its California fab; multiple future products taped out on its 300mm platform. No output, yield or customer-acceptance figures were disclosed. This early device stage does not replace InnoLight’s dated H2 2027 NPO ramp expectation.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-tower-npo-2026-11-03"]}]
- source_ids: ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"] → ["innolight-tower-npo-2026-11-03","innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"]
- reviewed_at: 2026-11-05 → 2026-11-07
- Updated: Ligent Technologies
3 field changes
- facts: [{"label":"High-speed product portfolio","value":"Company lists 400G, 800G and 1.6T pluggable transceiver families; the site does not establish shipment mix, customer qualification or production yield.","evidence":"vendor","as_of":"2026-09-22","source_ids":["ligent"]},{"label":"First-half 2026 scale","value":"Reuters reports RMB 5.39 billion revenue and RMB 661 million net profit for the six months ended June 30, up 27.9% and 29.7% year over year.","evidence":"reported","as_of":"2026-06-30","source_ids":["ligent-ipo-2026-09-22"]},{"label":"External light-source commercialization boundary","value":"Ligent reports initial commercialization of 100 mW ELSFP using self-developed CW lasers, while 400 mW ELSFP was scheduled for demonstration. It announced 6.4T/3.2T NPO and 12.8T TFLN XPO products; customer acceptance, shipment volume and component provenance remain unknown.","evidence":"vendor","as_of":"2026-09-07","source_ids":["ligent-cioe-2026-09-07"]}] → [{"label":"High-speed product portfolio","value":"Company lists 400G, 800G and 1.6T pluggable transceiver families; the site does not establish shipment mix, customer qualification or production yield.","evidence":"vendor","as_of":"2026-09-22","source_ids":["ligent"]},{"label":"First-half 2026 scale","value":"Reuters reports RMB 5.39 billion revenue and RMB 661 million net profit for the six months ended June 30, up 27.9% and 29.7% year over year.","evidence":"reported","as_of":"2026-06-30","source_ids":["ligent-ipo-2026-09-22"]},{"label":"External light-source commercialization boundary","value":"Ligent reports initial commercialization of 100 mW ELSFP using self-developed CW lasers, while 400 mW ELSFP was scheduled for demonstration. It announced 6.4T/3.2T NPO and 12.8T TFLN XPO products; customer acceptance, shipment volume and component provenance remain unknown.","evidence":"vendor","as_of":"2026-09-07","source_ids":["ligent-cioe-2026-09-07"]},{"label":"Qingdao chip expansion target","value":"Company-hosted reporting says Qingdao Wangtai cleanrooms are finished and MOCVD equipment is being installed. Trial and volume production are targeted for H1 2027; projected company-wide 2027 optical-chip capacity exceeds 170 million devices including existing lines, reportedly covered by customer demand. Neither realized output nor binding order terms were disclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["ligent-qingdao-capacity-2026-10-23"]}]
- source_ids: ["ligent","ligent-ipo-2026-09-22","ligent-cioe-2026-09-07"] → ["ligent-qingdao-capacity-2026-10-23","ligent","ligent-ipo-2026-09-22","ligent-cioe-2026-09-07"]
- reviewed_at: 2026-10-22 → 2026-11-07
- Updated: NAURA
3 field changes
- facts: [{"label":"Nine-month sales and delivery constraints","value":"NAURA reports RMB 31.467 billion revenue (+15.26%) and RMB 5.591 billion operating cash inflow (+317.84%). It says shortages in certain materials pressure some deliveries, while new-product customer validation and component upgrades contributed to lower margins. Affected inputs and order values are unspecified.","evidence":"vendor","as_of":"2026-09-30","source_ids":["naura-q3-2026-10-30"]},{"label":"Hybrid-bonding acquisition agreement","value":"NAURA Microelectronics signed an agreement to acquire 100% of Nuohe Jianwei for RMB 404.6 million. The target's hybrid-bonding systems were described as being in product validation and market introduction with repeat orders, but closing conditions remained and customer names, accepted production tools, throughput, yield and uptime were not disclosed.","evidence":"vendor","as_of":"2026-10-15","source_ids":["naura-nuohe-jianwei-acquisition-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reported RMB 20.161 billion of first-half revenue, up 24.90% year over year, and RMB 2.677 billion of R&D expense, up 28.87%.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Advanced packaging expansion","value":"Company reports that hybrid-bonding and high-end TSV electroplating tools were rolled out; the filing does not disclose named-customer acceptance, yield or production volume.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Atomic-layer-etch customer use","value":"Company says its ALE tool has achieved customer process use and an undated product disclosure says it entered a well-known customer's production line and completed process validation. The customer, process node, installed units, yield, uptime and production volume were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-primary","naura-ale-2026-09-24"]},{"label":"Near-term delivery constraint","value":"Company says tight delivery of materials from individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-2026-09-24"]}] → [{"label":"Nine-month sales and delivery constraints","value":"NAURA reports RMB 31.467 billion revenue (+15.26%) and RMB 5.591 billion operating cash inflow (+317.84%). It says shortages in certain materials pressure some deliveries, while new-product customer validation and component upgrades contributed to lower margins. Affected inputs and order values are unspecified.","evidence":"vendor","as_of":"2026-09-30","source_ids":["naura-q3-2026-10-30"]},{"label":"Hybrid-bonding acquisition agreement","value":"NAURA Microelectronics signed an agreement to acquire 100% of Nuohe Jianwei for RMB 404.6 million. The target's hybrid-bonding systems were described as being in product validation and market introduction with repeat orders, but closing conditions remained and customer names, accepted production tools, throughput, yield and uptime were not disclosed.","evidence":"vendor","as_of":"2026-10-15","source_ids":["naura-nuohe-jianwei-acquisition-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reported RMB 20.161 billion of first-half revenue, up 24.90% year over year, and RMB 2.677 billion of R&D expense, up 28.87%.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Advanced packaging expansion","value":"Company reports that hybrid-bonding and high-end TSV electroplating tools were rolled out; the filing does not disclose named-customer acceptance, yield or production volume.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Atomic-layer-etch customer use","value":"Company says its ALE tool has achieved customer process use and an undated product disclosure says it entered a well-known customer's production line and completed process validation. The customer, process node, installed units, yield, uptime and production volume were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-primary","naura-ale-2026-09-24"]},{"label":"Near-term delivery constraint","value":"Company says tight delivery of materials from individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-2026-09-24"]},{"label":"Microwave ECR product launch","value":"NAURA announced its first 12-inch Acme Helix430 microwave ECR plasma etcher, complementing ICP/CCP tools and targeting low-damage advanced-node etching. Claimed process validation is not accompanied by named-customer qualification, accepted tool counts, yield, throughput or uptime.","evidence":"vendor","as_of":"2026-10-29","source_ids":["naura-ecr-launch-2026-10-29"]}]
- source_ids: ["naura-q3-2026-10-30","naura","naura-2026-h1-report","naura-ale-primary","naura-ale-2026-09-24","naura-nuohe-jianwei-acquisition-2026-10-16"] → ["naura-ecr-launch-2026-10-29","naura-q3-2026-10-30","naura","naura-2026-h1-report","naura-ale-primary","naura-ale-2026-09-24","naura-nuohe-jianwei-acquisition-2026-10-16"]
- reviewed_at: 2026-11-01 → 2026-11-07
- Updated: RoboTechnik
3 field changes
- facts: [{"label":"Silicon-photonics backlog","value":"Company reports RMB 2.0 billion of SiPh assembly and testing equipment backlog at June 30, without customer, margin or acceptance details.","evidence":"vendor","as_of":"2026-06-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Silicon-photonics revenue mix","value":"SiPh assembly and test equipment generated RMB 439.1 million in 2025 and 65.8% of group revenue in the first four months of 2026 after the ficonTEC acquisition.","evidence":"vendor","as_of":"2026-04-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Offering plan","value":"At the maximum offer price, the prospectus estimates HKD 4.961 billion of net proceeds, with 20% planned for R&D and 40% for capacity; final proceeds and execution remain pending.","evidence":"primary","as_of":"2026-09-21","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20"]},{"label":"Final base offering terms","value":"The September 28 allotment filing sets 11,876,000 H shares at HKD 436 each, with HKD 5,177.9 million gross and HKD 4,961.3 million net proceeds before any over-allotment exercise. The offer-size adjustment option was not exercised; over-allocation was 1,781,400 shares. This resolves base-offering pricing, not realized equipment capacity or spending.","evidence":"primary","as_of":"2026-09-28","source_ids":["robotechnik-allotment-2026-09-28"]}] → [{"label":"Silicon-photonics backlog","value":"Company reports RMB 2.0 billion of SiPh assembly and testing equipment backlog at June 30, without customer, margin or acceptance details.","evidence":"vendor","as_of":"2026-06-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Silicon-photonics revenue mix","value":"SiPh assembly and test equipment generated RMB 439.1 million in 2025 and 65.8% of group revenue in the first four months of 2026 after the ficonTEC acquisition.","evidence":"vendor","as_of":"2026-04-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Offering plan","value":"At the maximum offer price, the prospectus estimates HKD 4.961 billion of net proceeds, with 20% planned for R&D and 40% for capacity; final proceeds and execution remain pending.","evidence":"primary","as_of":"2026-09-21","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20"]},{"label":"Final base offering terms","value":"The September 28 allotment filing sets 11,876,000 H shares at HKD 436 each, with HKD 5,177.9 million gross and HKD 4,961.3 million net proceeds before any over-allotment exercise. The offer-size adjustment option was not exercised; over-allocation was 1,781,400 shares. This resolves base-offering pricing, not realized equipment capacity or spending.","evidence":"primary","as_of":"2026-09-28","source_ids":["robotechnik-allotment-2026-09-28"]},{"label":"Equipment backlog at disclosure","value":"The unaudited quarterly filing reports RMB 2.359 billion of optoelectronic and semiconductor orders not yet recognized as revenue at disclosure. This broader category is not directly comparable with June SiPh backlog. It reports scaled deliveries and revenue recognition for coupling/test equipment, without unit counts or customer-level acceptance metrics.","evidence":"vendor","as_of":"2026-10-28","source_ids":["robotechnik-q3-backlog-2026-10-28"]}]
- source_ids: ["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","robotechnik-allotment-2026-09-28"] → ["robotechnik-q3-backlog-2026-10-28","robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","robotechnik-allotment-2026-09-28"]
- reviewed_at: 2026-10-23 → 2026-11-07
- Added: InnoLight NPO device enters early production at Tower California fab
- Added: Ligent describes Qingdao optical-chip installation and 2027 capacity targets
- Added: NAURA announces Acme Helix430 microwave ECR etcher
- Added: RoboTechnik reports RMB 2.359 billion equipment backlog and further overseas capacity plans
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-11-06 → 2026-11-07
- coverage: November 6 partial sweep: 12 priority-1 companies, four oldest remaining profiles and Hwatsing screened; YMTC and Hwatsing substantively updated. Other profile dates unchanged. → November 7 partial sweep: 12 priority-1 companies and five oldest remaining profiles screened; InnoLight, NAURA, Ligent and RoboTechnik substantively updated. Other review dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-07","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","guowang","robotechnik"],"summary":"Screened priority companies and five oldest remaining profiles with English/Chinese searches overlapping October 30–November 7, plus older-source follow-ups. Read RoboTechnik’s quarterly filing, Ligent’s company-hosted October 23 report and CLS dispatch, NAURA’s company-authored announcement reprint and official news index, and Tower’s November 3 release. Only InnoLight, NAURA, Ligent and RoboTechnik profile dates advanced. RoboTechnik backlog is as of disclosure, covers optoelectronic/semiconductor business and is not directly comparable to June SiPh backlog. Ligent company-wide capacity is a target including existing lines; CLS separately described over 100 million devices at the new factory and a mid-2027 start. Neither establishes output. Tower’s early device production is preserved alongside the existing later NPO ramp forecast. Other searches chiefly returned existing or secondary coverage. RSLaser/Guowang qualified-production evidence, T-Head software/code/license inspection, CXMT prototype primary verification and RoboTechnik October 29 IR remain follow-ups. MLPerf Training v6.0 and Inference v6.1 latest commit checks were unchanged at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and 4bb63cd28eb136be221bc801a9e38c66c8a7553f; artifacts were not exhaustively audited. The retrieved InferenceX landing page displayed an unofficial-deployment warning; its new content was not accepted as verified benchmark evidence. Resolve dashboard provenance against original runs before adding results. No new matched frontier end-to-end benchmark verified; benchmark review date and historical studies unchanged. Coverage remains partial for private deployments, tenders, unindexed disclosures, customer acceptance, yields, energy, reliability and component provenance. Historical reports and forecasts preserved.","source_ids":["robotechnik-q3-backlog-2026-10-28","ligent-qingdao-capacity-2026-10-23","ligent-qingdao-cls-2026-10-23","naura-ecr-launch-2026-10-29","innolight-tower-npo-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: InnoLight NPO device enters early production at Tower California fab
- Added: Ligent describes Qingdao optical-chip installation and 2027 capacity targets
- Added: CLS reports Qingdao optical-chip factory installation and planned capacity
- Added: NAURA announces Acme Helix430 microwave ECR etcher
- Added: RoboTechnik reports RMB 2.359 billion equipment backlog and further overseas capacity plans
2026-11-06dc148a2 ↗
research: record YMTC sourcing target and Hwatsing acceptance claims
- Updated: Hwatsing
3 field changes
- facts: [{"label":"Placement share registration","value":"Hwatsing reports completing registration of the placement's 16,080,508 new shares on October 15, with net proceeds of RMB 3.7731166749 billion. This is the same financing already recorded, not additional funding or evidence of completed production capacity.","evidence":"primary","as_of":"2026-10-15","source_ids":["hwatsing-share-registration-2026-10-17"]},{"label":"Equipment and capacity financing","value":"Hwatsing completed a private placement of 16,080,508 shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D; schedules, installed capacity and realized output remain future work.","evidence":"primary","as_of":"2026-09-30","source_ids":["hwatsing-placement-2026-09-30"]},{"label":"Process scope","value":"The vendor describes automated endpoint detection and multiple polishing modes for large-format panel-level packaging; customer and production metrics were not disclosed.","evidence":"reported","as_of":"2026-09-09","source_ids":["hwatsing-panel-cmp-2026-09-09"]}] → [{"label":"Placement share registration","value":"Hwatsing reports completing registration of the placement's 16,080,508 new shares on October 15, with net proceeds of RMB 3.7731166749 billion. This is the same financing already recorded, not additional funding or evidence of completed production capacity.","evidence":"primary","as_of":"2026-10-15","source_ids":["hwatsing-share-registration-2026-10-17"]},{"label":"Equipment and capacity financing","value":"Hwatsing completed a private placement of 16,080,508 shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D; schedules, installed capacity and realized output remain future work.","evidence":"primary","as_of":"2026-09-30","source_ids":["hwatsing-placement-2026-09-30"]},{"label":"Process scope","value":"The vendor describes automated endpoint detection and multiple polishing modes for large-format panel-level packaging; customer and production metrics were not disclosed.","evidence":"reported","as_of":"2026-09-09","source_ids":["hwatsing-panel-cmp-2026-09-09"]},{"label":"Acceptance and repeat-order claims","value":"At its November 5 results meeting, management reportedly said multiple thinning tools had passed customer acceptance and 12-inch high-current ion implanters had batch deliveries and repeat orders. Medium-current models remained in development and customer validation; customers, units and operating metrics were not disclosed.","evidence":"reported","as_of":"2026-11-05","source_ids":["hwatsing-acceptance-2026-11-05"]}]
- source_ids: ["hwatsing","hwatsing-panel-cmp-2026-09-09","hwatsing-placement-2026-09-30","hwatsing-share-registration-2026-10-17"] → ["hwatsing","hwatsing-panel-cmp-2026-09-09","hwatsing-placement-2026-09-30","hwatsing-share-registration-2026-10-17","hwatsing-acceptance-2026-11-05"]
- reviewed_at: 2026-10-27 → 2026-11-06
- Updated: YMTC
3 field changes
- facts: [{"label":"Capacity expansion","value":"Reuters reported two additional factories planned beyond one due for completion in 2026; completed capacity, qualified output and domestic equipment provenance remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Equipment diversification","value":"The Wire China reported that YMTC replaced blocked U.S. chipmaking tools with Dutch and Japanese equipment; suppliers, models, domestic share and complete-flow provenance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Licensed process IP","value":"The Wire China reported that at least one YMTC innovation was licensed to Samsung; the patent, terms, production use and economic contribution were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Pricing and product mix","value":"The Wire China reportedly quoted one unnamed employee saying YMTC had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. Affected products, price changes, capacity allocation and corporate guidance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06"]},{"label":"Global shipment scale","value":"Counterpoint estimated YMTC held 14% of global NAND shipment volume in the second quarter of 2026, ranking third, but ranked fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low enterprise-SSD share. A later Counterpoint article described a 16.5% current shipment share without publicly reconciling the measurement basis or revision with the earlier 14% estimate.","evidence":"measured","as_of":"2026-06-30","source_ids":["counterpoint-ymtc-q2-shipments-2026-08-12","counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"DRAM research and testing","value":"Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month and explicitly said the device had no mass-production plan. A faster 9,600 MT/s version remained a development target for Fab 3.","evidence":"reported","as_of":"2026-10-13","source_ids":["counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"Proposed STAR Market financing","value":"The Shanghai Stock Exchange accepted YMTC's application for a proposed RMB 33 billion STAR Market raise; the project was under inquiry and did not establish completed financing or installed capacity.","evidence":"primary","as_of":"2026-08-21","source_ids":["ymtc-star-market-ipo-2026-08-21"]}] → [{"label":"Capacity expansion","value":"Reuters reported two additional factories planned beyond one due for completion in 2026; completed capacity, qualified output and domestic equipment provenance remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Equipment diversification","value":"The Wire China reported that YMTC replaced blocked U.S. chipmaking tools with Dutch and Japanese equipment; suppliers, models, domestic share and complete-flow provenance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Licensed process IP","value":"The Wire China reported that at least one YMTC innovation was licensed to Samsung; the patent, terms, production use and economic contribution were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04"]},{"label":"Pricing and product mix","value":"The Wire China reportedly quoted one unnamed employee saying YMTC had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. Affected products, price changes, capacity allocation and corporate guidance were not disclosed.","evidence":"reported","as_of":"2026-10-04","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06"]},{"label":"Global shipment scale","value":"Counterpoint estimated YMTC held 14% of global NAND shipment volume in the second quarter of 2026, ranking third, but ranked fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low enterprise-SSD share. A later Counterpoint article described a 16.5% current shipment share without publicly reconciling the measurement basis or revision with the earlier 14% estimate.","evidence":"measured","as_of":"2026-06-30","source_ids":["counterpoint-ymtc-q2-shipments-2026-08-12","counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"DRAM research and testing","value":"Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month and explicitly said the device had no mass-production plan. A faster 9,600 MT/s version remained a development target for Fab 3.","evidence":"reported","as_of":"2026-10-13","source_ids":["counterpoint-ymtc-ipo-dram-2026-10-13"]},{"label":"Proposed STAR Market financing","value":"The Shanghai Stock Exchange accepted YMTC's application for a proposed RMB 33 billion STAR Market raise; the project was under inquiry and did not establish completed financing or installed capacity.","evidence":"primary","as_of":"2026-08-21","source_ids":["ymtc-star-market-ipo-2026-08-21"]},{"label":"Fourth-fab equipment target","value":"Reuters reports a target of at least 60% domestic equipment, with an ambition of 70%, for the planned fourth fab. Supplier talks remain preliminary, with no formal orders.","evidence":"reported","as_of":"2026-11-06","source_ids":["ymtc-domestic-tools-2026-11-06"]}]
- source_ids: ["ymtc","china-memory-expansion-reuters-2026-07-24","ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","counterpoint-ymtc-q2-shipments-2026-08-12","ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-ipo-dram-2026-10-13"] → ["ymtc","china-memory-expansion-reuters-2026-07-24","ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","counterpoint-ymtc-q2-shipments-2026-08-12","ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-ipo-dram-2026-10-13","ymtc-domestic-tools-2026-11-06"]
- reviewed_at: 2026-10-16 → 2026-11-06
- Added: Hwatsing reports thinning acceptance and ion-implantation repeat orders
- Added: YMTC reportedly targets higher domestic equipment share
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-05 → 2026-11-06
- last_automated_run_at: 2026-11-05 → 2026-11-06
- coverage: November 5 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Skyverse and InnoLight substantively updated. Other review dates unchanged. → November 6 partial sweep: 12 priority-1 companies, four oldest remaining profiles and Hwatsing screened; YMTC and Hwatsing substantively updated. Other profile dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-06","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","ligent","robotechnik","hwatsing"],"summary":"Screened priority companies and four oldest remaining profiles with overlapping English/Chinese searches covering October 29–November 6; additionally reviewed Hwatsing. Read original Reuters reporting and the CLS results-meeting report. Only YMTC and Hwatsing profile dates advanced. Hwatsing financial figures in the article were not adopted; original filing verification remains necessary. RoboTechnik quarterly backlog and October 29 investor record, Ligent capacity-locking coverage and T-Head software release/code/license inspection remain follow-ups. Lithography searches returned older secondary coverage without new qualified-production evidence. Other priority searches mainly reiterated existing reports. GitHub confirmed unchanged MLPerf Training v6.0 and Inference v6.1 heads; read InferenceX landing coverage. No new matched frontier end-to-end comparison verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["ymtc-domestic-tools-2026-11-06","hwatsing-acceptance-2026-11-05","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Hwatsing reports thinning acceptance and ion-implantation repeat orders
- Added: YMTC reportedly targets higher domestic equipment share
2026-11-05a626d71 ↗
research: add Skyverse shipment and InnoLight NPO supply evidence
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]},{"label":"NPO roadmap and component constraints","value":"Company expects NPO ramp-up in H2 2027 and greater demand in 2028. Product manufacturing yields and new upstream materials, including higher-power continuous-wave light sources and electronic chips, still require yield ramp-up. It says secured DSP supply is close to order needs, without quantities or supplier identities.","evidence":"vendor","as_of":"2026-11-03","source_ids":["innolight-ir-2026-11-03"]}]
- source_ids: ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30"] → ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30","innolight-ir-2026-11-03"]
- reviewed_at: 2026-10-30 → 2026-11-05
- Updated: Skyverse
3 field changes
- facts: [{"label":"Customer validation and sales","value":"Company says 11 of its 13 equipment families passed leading domestic customer validation and generated sales. Mature-process bright-field and X-ray TSV copper-void tools passed validation; HBM wafer-flatness tools remained under evaluation, and electron-beam critical-dimension tools had shipped for validation. Customers, nodes, unit volumes and operating metrics were undisclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["skyverse-ir-2026-09-11"]}] → [{"label":"Customer validation and sales","value":"Company says 11 of its 13 equipment families passed leading domestic customer validation and generated sales. Mature-process bright-field and X-ray TSV copper-void tools passed validation; HBM wafer-flatness tools remained under evaluation, and electron-beam critical-dimension tools had shipped for validation. Customers, nodes, unit volumes and operating metrics were undisclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["skyverse-ir-2026-09-11"]},{"label":"Cumulative shipment milestone","value":"Company reports dispatching its 2,000th IC quality-control tool to an unnamed domestic chipmaker under a signed order. Shipment does not establish customer acceptance; equipment mix, process nodes, throughput and yield contribution remain undisclosed.","evidence":"vendor","as_of":"2026-10-31","source_ids":["skyverse-2000th-shipment-2026-10-31"]}]
- source_ids: ["skyverse","skyverse-ir-2026-09-11"] → ["skyverse","skyverse-ir-2026-09-11","skyverse-2000th-shipment-2026-10-31"]
- reviewed_at: 2026-10-21 → 2026-11-05
- Added: InnoLight outlines NPO yield requirements and supply outlook
- Added: Skyverse announces its 2,000th equipment shipment
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-04 → 2026-11-05
- last_automated_run_at: 2026-11-04 → 2026-11-05
- coverage: November 4 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Eoptolink, Huawei, Iluvatar and Moore Threads substantively updated. Other review dates unchanged. → November 5 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Skyverse and InnoLight substantively updated. Other review dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-05","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","thead","skyverse","ligent"],"summary":"Screened priority companies and the four oldest remaining profiles with overlapping English/Chinese searches covering October 28–November 5. Read Skyverse's shipment filing and InnoLight's November 3 investor-call PDF; only these two profile review dates advanced. InnoLight's primary disclosure resolves part of the earlier NPO/DSP follow-up, while delivered component volumes remain unknown. T-Head software-release and open-source claims still require original release, code and license inspection. Corrected the Ligent search to 纳真科技; indexed news and capital filings did not establish a new operational milestone. RSLaser searches did not substantiate a new qualification event. Other equipment, accelerator and memory queries mostly returned prior disclosures or secondary material; no new qualifying production or performance claim accepted. GitHub confirmed unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Reviewed InferenceX landing coverage; no new matched frontier end-to-end result verified. Individual artifacts were not exhaustively audited; existing Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["skyverse-2000th-shipment-2026-10-31","innolight-ir-2026-11-03","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: InnoLight outlines NPO yield requirements and supply outlook
- Added: Skyverse announces its 2,000th equipment shipment
2026-11-04b489efd ↗
research: add shipment, cash-flow and industrial adoption evidence
- Updated: Eoptolink
3 field changes
- facts: [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"1.6T volume-ramp disclosure","value":"Company says 800G and 1.6T are its main delivered products in 2026 and 1.6T has entered a continuing volume ramp. Faster future shipments remain an expectation; unit counts, yields, utilization and named customer acceptance are undisclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["eoptolink-operations-2026-10-23"]},{"label":"Nine-month growth and cash-flow pressure","value":"The unaudited nine-month filing reports revenue of RMB 31.850 billion, up 92.97%. Operating cash flow fell 68.42% to RMB 1.464 billion, attributed to increased material payments and taxes. September inventory reached RMB 17.413 billion, up 140.70% from year-end, for existing and expected orders and stocking cycles. The filing does not quantify qualified capacity or component provenance.","evidence":"vendor","as_of":"2026-09-30","source_ids":["eoptolink-q3-2026-10-30"]}] → [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"1.6T volume-ramp disclosure","value":"Company says 800G and 1.6T are its main delivered products in 2026 and 1.6T has entered a continuing volume ramp. Faster future shipments remain an expectation; unit counts, yields, utilization and named customer acceptance are undisclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["eoptolink-operations-2026-10-23"]},{"label":"Nine-month growth and cash-flow pressure","value":"The unaudited nine-month filing reports revenue of RMB 31.850 billion, up 92.97%. Operating cash flow fell 68.42% to RMB 1.464 billion, attributed to increased material payments and taxes. September inventory reached RMB 17.413 billion, up 140.70% from year-end, for existing and expected orders and stocking cycles. The filing does not quantify qualified capacity or component provenance.","evidence":"vendor","as_of":"2026-09-30","source_ids":["eoptolink-q3-2026-10-30"]},{"label":"Shipment timing and material constraints","value":"Company says late-September Thai floods delayed logistics and revenue recognition, with affected shipments recognized in October; production and goods were unaffected. Materials remain broadly tight, with longer stocking cycles. Quantities and component provenance are undisclosed.","evidence":"vendor","as_of":"2026-10-29","source_ids":["eoptolink-ir-2026-10-29"]}]
- source_ids: ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22","eoptolink-operations-2026-10-23","eoptolink-q3-2026-10-30"] → ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22","eoptolink-operations-2026-10-23","eoptolink-q3-2026-10-30","eoptolink-ir-2026-10-29"]
- reviewed_at: 2026-10-30 → 2026-11-04
- Updated: Huawei
3 field changes
- facts: [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]},{"label":"Independent Ascend 950PR kernel measurements","value":"SemiAnalysis tested a 950PR_9579 with 28 Cube cores and 128 GiB HBM. These measurements do not establish end-to-end model parity.","evidence":"measured","as_of":"2026-10-21","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"label":"DeepInfer public inference engine","value":"DeepSeek released an Apache-2.0 offline engine for V4.1-Flash supporting Ascend 950 with CANN 9.2 and NVIDIA SM100. The README excludes the KV-cache service, complete disaggregation and replica load balancing; batch-invariant inference and deterministic sampling are NVIDIA-only. Its published performance plots use NVIDIA hardware, not Ascend. Code availability does not establish measured production parity.","evidence":"primary","as_of":"2026-10-20","source_ids":["deepinfer-release-2026-10-20"]}] → [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]},{"label":"Independent Ascend 950PR kernel measurements","value":"SemiAnalysis tested a 950PR_9579 with 28 Cube cores and 128 GiB HBM. These measurements do not establish end-to-end model parity.","evidence":"measured","as_of":"2026-10-21","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"label":"DeepInfer public inference engine","value":"DeepSeek released an Apache-2.0 offline engine for V4.1-Flash supporting Ascend 950 with CANN 9.2 and NVIDIA SM100. The README excludes the KV-cache service, complete disaggregation and replica load balancing; batch-invariant inference and deterministic sampling are NVIDIA-only. Its published performance plots use NVIDIA hardware, not Ascend. Code availability does not establish measured production parity.","evidence":"primary","as_of":"2026-10-20","source_ids":["deepinfer-release-2026-10-20"]},{"label":"Named steel-industry adoption","value":"Huawei says Nanjing Iron and Steel uses Ascend, Kunpeng, Huawei Cloud Stack and Pangu for its Yuanye system, with 25 business agents online by October. Accelerator counts, utilization, model throughput and independent operating validation are undisclosed.","evidence":"vendor","as_of":"2026-10-27","source_ids":["huawei-nisco-2026-10-27"]}]
- source_ids: ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19","semianalysis-ascend-950pr-2026-10-21","deepinfer-release-2026-10-20"] → ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19","semianalysis-ascend-950pr-2026-10-21","deepinfer-release-2026-10-20","huawei-nisco-2026-10-27"]
- reviewed_at: 2026-10-23 → 2026-11-04
- Updated: Iluvatar CoreX
3 field changes
- facts: [{"label":"Product revenue mix","value":"Company reports RMB 654.2 million of inference-product revenue, up 651.8%, and RMB 261.9 million of training-product revenue, up 38.0%; units, average prices and model mix are undisclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]},{"label":"Delivery and supply qualification","value":"AI-computing-solutions revenue fell 68.1% to RMB 13.6 million; the company cited strategic focus plus supply-resource coordination and delivery-schedule adjustments, without establishing a specific chip shortage or customer cancellation.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]},{"label":"Economic qualification","value":"Total gross margin fell to 17.2% from 50.1%. Net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]}] → [{"label":"Product revenue mix","value":"Company reports RMB 654.2 million of inference-product revenue, up 651.8%, and RMB 261.9 million of training-product revenue, up 38.0%; units, average prices and model mix are undisclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]},{"label":"Delivery and supply qualification","value":"AI-computing-solutions revenue fell 68.1% to RMB 13.6 million; the company cited strategic focus plus supply-resource coordination and delivery-schedule adjustments, without establishing a specific chip shortage or customer cancellation.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]},{"label":"Economic qualification","value":"Total gross margin fell to 17.2% from 50.1%. Net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion.","evidence":"vendor","as_of":"2026-06-30","source_ids":["iluvatar-2026-h1-report"]},{"label":"Reported ten-month shipments","value":"Sina relays a company announcement of more than 80,000 chips shipped through October 31. The figure is unaudited, subject to adjustment and explicitly not a revenue measure. Product mix, acceptance, utilization and component provenance remain unverified.","evidence":"reported","as_of":"2026-10-31","source_ids":["iluvatar-shipments-2026-11-02"]}]
- source_ids: ["iluvatar","iluvatar-2026-h1-report"] → ["iluvatar","iluvatar-2026-h1-report","iluvatar-shipments-2026-11-02"]
- reviewed_at: 2026-10-12 → 2026-11-04
- Updated: Moore Threads
3 field changes
- facts: [{"label":"Wuxi operating deployment","value":"The operator reported all 2,000 MTT S5000 cards in phase one of the Wuxi Huishan center were operating and aggregate demand was near full load, with more than 20 customers testing or using the cluster. Workload-level utilization, reliability, power and performance were undisclosed.","evidence":"reported","as_of":"2026-08-25","source_ids":["moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reports RMB 1.736 billion of first-half revenue, up 147.42% year over year, and an attributable net loss of RMB 11.56 million.","evidence":"vendor","as_of":"2026-06-30","source_ids":["moore-threads-2026-h1-report"]},{"label":"Vendor-reported KUAE training metrics","value":"Company reports Dense-model MFU up to 60%, MoE-model MFU up to 40%, 95% linear scaling efficiency and effective training time above 90%; workload, power and matched comparator details were not disclosed in the filing.","evidence":"vendor","as_of":"2026-06-30","source_ids":["moore-threads-2026-h1-report"]},{"label":"JD Cloud cluster plan","value":"JD Cloud announced a plan to build a 100,000-GPU cluster based on Moore Threads; no delivery schedule, completed card count, topology or operational results were disclosed.","evidence":"vendor","as_of":"2026-09-09","source_ids":["moore-threads-jd-cloud-2026-09-09"]}] → [{"label":"Wuxi operating deployment","value":"The operator reported all 2,000 MTT S5000 cards in phase one of the Wuxi Huishan center were operating and aggregate demand was near full load, with more than 20 customers testing or using the cluster. Workload-level utilization, reliability, power and performance were undisclosed.","evidence":"reported","as_of":"2026-08-25","source_ids":["moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reports RMB 1.736 billion of first-half revenue, up 147.42% year over year, and an attributable net loss of RMB 11.56 million.","evidence":"vendor","as_of":"2026-06-30","source_ids":["moore-threads-2026-h1-report"]},{"label":"Vendor-reported KUAE training metrics","value":"Company reports Dense-model MFU up to 60%, MoE-model MFU up to 40%, 95% linear scaling efficiency and effective training time above 90%; workload, power and matched comparator details were not disclosed in the filing.","evidence":"vendor","as_of":"2026-06-30","source_ids":["moore-threads-2026-h1-report"]},{"label":"JD Cloud cluster plan","value":"JD Cloud announced a plan to build a 100,000-GPU cluster based on Moore Threads; no delivery schedule, completed card count, topology or operational results were disclosed.","evidence":"vendor","as_of":"2026-09-09","source_ids":["moore-threads-jd-cloud-2026-09-09"]},{"label":"Third-quarter commercialization and cash use","value":"Unaudited Q3 revenue was RMB 1.552 billion and attributable profit RMB 49.26 million. Nine-month revenue was RMB 3.289 billion; profit excluding nonrecurring items remained negative at RMB 130.75 million and operating cash outflow reached RMB 3.309 billion.","evidence":"vendor","as_of":"2026-09-30","source_ids":["moore-threads-q3-2026-10-29"]}]
- source_ids: ["mthreads","moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16"] → ["mthreads","moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","moore-threads-q3-2026-10-29"]
- reviewed_at: 2026-10-17 → 2026-11-04
- Added: Eoptolink qualifies delayed shipments and material supply
- Added: Huawei reports Yuanye deployment at Nanjing Iron and Steel
- Added: Iluvatar reports more than 80,000 chips shipped in ten months
- Added: Moore Threads reports quarterly profit alongside substantial cash outflow
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-03 → 2026-11-04
- last_automated_run_at: 2026-11-03 → 2026-11-04
- coverage: November 3 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Enflame substantively updated. Search-only profiles retain their review dates. → November 4 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Eoptolink, Huawei, Iluvatar and Moore Threads substantively updated. Other review dates unchanged.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-04","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened priority companies and four oldest remaining profiles using overlapping English/Chinese searches covering October 27–November 4. Read Eoptolink's investor-call PDF, Moore Threads' unaudited quarterly filing and Huawei's named industrial-deployment release; added four events including Iluvatar's reported shipment update. Only these four profile review dates advanced. Iluvatar's original exchange shipment filing remains to be retrieved; media H200-performance claims were not promoted to measured evidence. InnoLight NPO/DSP commentary and T-Head software-release details remain follow-ups. Equipment, memory and lithography screening mostly returned previously recorded disclosures or older secondary coverage; no new qualified-tool or yield milestone accepted. Reviewed MLPerf Training v6.0 listing and InferenceX landing coverage; Inference v6.1 web retrieval failed, while GitHub confirmed unchanged head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end result verified; individual artifacts were not exhaustively audited and existing Ascend microbenchmarks remain distinct. Coverage is partial for private deployments, tenders, unindexed filings, customer acceptance, production yields, reliability, energy and component provenance. Historical reports and forecasts preserved.","source_ids":["eoptolink-ir-2026-10-29","moore-threads-q3-2026-10-29","huawei-nisco-2026-10-27","iluvatar-shipments-2026-11-02","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Investor relations activity record 2026-006
- Added: Yuanye steel model deployed at Nanjing Iron and Steel
- Added: Iluvatar voluntary shipment update: more than 80,000 chips in ten months
- Added: 2026 third-quarter report (unaudited)
2026-11-0351017ea ↗
research: qualify Enflame long-context inference claims
- Updated: Enflame
3 field changes
- facts: [{"label":"System partner and corrected model name","value":"Enflame's October 16 release, republished October 18, identifies the model as ESL64-O (letter O) and names ZTE as co-developer. Its claim of supporting clusters above 10,000 cards describes design capability, not a measured or accepted deployment. See the appended correction to the earlier exhibition record.","evidence":"vendor","as_of":"2026-10-16","source_ids":["enflame-ciif-release-2026-10-16"]},{"label":"ESL64-O supernode","value":"Enflame displayed an ESL64-O supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}] → [{"label":"Long-context inference software claim","value":"In its October 29 conference release, Enflame describes operator fusion, memory optimization, scheduling, hierarchical KV Cache and prefill/decode separation, claiming stable high-throughput inference at million-token context lengths. It supplies no model, precision, accelerator count, latency, throughput, power or reproducible test configuration. This is a vendor capability claim, not a verified operating benchmark.","evidence":"vendor","as_of":"2026-10-29","source_ids":["enflame-tencent-inference-2026-10-29"]},{"label":"System partner and corrected model name","value":"Enflame's October 16 release, republished October 18, identifies the model as ESL64-O (letter O) and names ZTE as co-developer. Its claim of supporting clusters above 10,000 cards describes design capability, not a measured or accepted deployment. See the appended correction to the earlier exhibition record.","evidence":"vendor","as_of":"2026-10-16","source_ids":["enflame-ciif-release-2026-10-16"]},{"label":"ESL64-O supernode","value":"Enflame displayed an ESL64-O supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}]
- source_ids: ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12","enflame-ciif-release-2026-10-16"] → ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12","enflame-ciif-release-2026-10-16","enflame-tencent-inference-2026-10-29"]
- reviewed_at: 2026-10-21 → 2026-11-03
- Added: Enflame describes long-context inference software at Tencent conference
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-02 → 2026-11-03
- last_automated_run_at: 2026-11-02 → 2026-11-03
- coverage: November 2 daily sweep and October 26–November 1 weekly synthesis completed; underlying research remains partial. The weekly report covers eight selected events across seven companies and does not advance profile review dates. → November 3 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; Enflame substantively updated. Search-only profiles retain their review dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]},{"date":"2026-11-03","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","moore-threads"],"summary":"Screened 12 priority-1 companies and four oldest remaining profiles using grouped English/Chinese searches over the October 26–November 3 overlap. Read the Enflame-origin October 29 conference release republished by ijiwei and added narrowly attributed inference-software evidence; only Enflame profile review date advanced. The source uses an October 29 release dateline; the relay displays a relative posting timestamp. Iluvatar reporting names Shanghai INESA as the July H200-comparison tester, but no original test report or full configurations were obtained; no ratio was promoted to measured evidence. T-Head and Moore Threads queries returned older material without a new verified event. Lithography searches returned older secondary reports, not new acceptance, yield or production evidence. Optics broker commentary about DSP supply, delayed deliveries and NPO timing remains a primary-filing follow-up. CXMT search results reiterated already-recorded sampling and financial disclosures. Reviewed MLPerf Training landing coverage and InferenceX; MLPerf Inference web retrieval failed, but GitHub confirmed unchanged v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No new matched frontier end-to-end comparison verified; artifacts were not exhaustively audited. Existing independent Ascend kernel measurements remain distinct. Coverage remains partial for private deployments, tenders, unindexed filings, customer acceptance, unit volumes, yields, reliability, energy and component provenance. Historical reports and dated forecasts preserved.","source_ids":["enflame-tencent-inference-2026-10-29","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Enflame Tencent conference release: inference software and long-context claims
2026-11-027fcf280 ↗
briefing: synthesize October 26–November 1 evidence
- Added: Memory constraints and capacity investment frame domestic compute expansion
- Updated: data/research-state.yaml
2 field changes
- coverage: November 2 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; CXMT and MetaX substantively updated. Search-only profiles retain their actual review dates. → November 2 daily sweep and October 26–November 1 weekly synthesis completed; underlying research remains partial. The weekly report covers eight selected events across seven companies and does not advance profile review dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-02","status":"partial","companies":["cxmt","metax","cambricon","jcet","naura","innolight","eoptolink"],"summary":"Prepared the missing October 26–November 1 weekly briefing with the weekly helper as of November 2. Reviewed thirteen in-period archive records and synthesized eight selected events. Reread Reuters relay, Cambricon clarification, JCET investment reporting, NAURA and optical filings; ran overlapping English/Chinese gap searches. Repeat MetaX and CXMT sampling retrieval failed, so retained the separately recorded same-day primary-source review and disclosed this limitation. Excluded evolving upstream PR details from period-end claims. Publication-day benchmark checks are labeled post-period. No new profile review dates or forecasts; all prior editions preserved. Closed-period report complete, research coverage partial.","source_ids":["cambricon-shipment-clarification-2026-10-26","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-rdimm-sampling-2026-10-31","eoptolink-q3-2026-10-30","inferencex","innolight-q3-2026-10-30","jcet-packaging-investment-2026-10-28","metax-rlinf-validation-2026-10-26","mlperf-inference-6-1","mlperf-training","naura-q3-2026-10-30"]}]
2026-11-02869cc33 ↗
research: clarify CXMT sampling and MetaX RLinf validation
- Updated: data/benchmarks.yaml
3 field changes
- assessment: Follow-up through October 27 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through November 2 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-27 → 2026-11-02
- studies: [{"title":"ubenchX core memory-access profiles","evidence":"measured","date":"Reviewed 2026-10-27; publication date unverified","finding":"SemiAnalysis publishes scalar pointer-chase memory profiles for Ascend 950PR_9579 (128 MiB allocation, Cube and Vector controllers) and B200. Values measure differences between matched-address profiles, not direct core-to-core latency or physical distance. Ascend SYS_CNT ticks and B200 clock64 cycles are different units and cannot support a cross-device latency ratio. Concurrent launches lack a grid-wide starting barrier; fine grouping is less repeatable and does not establish physical cache sharing. These are hardware microbenchmarks, not end-to-end AI performance; the Observatory did not rerun them.","source_ids":["inferencex-core-memory-profiles"]},{"title":"DeepInfer published performance scope","evidence":"vendor","date":"2026-10-20","finding":"DeepSeek publishes prefill and decode throughput/latency plots using 4 and 64 NVIDIA SM100 GPUs respectively, each within one scale-up domain. Ascend 950 is a supported software target, but these plots are not Ascend measurements or a matched cross-hardware comparison. The offline release excludes complete service infrastructure. No benchmark was rerun by the Observatory.","source_ids":["deepinfer-release-2026-10-20"]},{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}] → [{"title":"MetaX RLinf embodied training","evidence":"vendor","date":"2026-10-26","finding":"MetaX reports 1,024-GPU Pi0.5/RoboTwin SFT: 20,000 steps in 32 hours, 2.75 samples/s per GPU versus 2.8 at 512 GPUs, using RoCE and GPFS. Its separate greater-than-97% scaling claim covers ManiSkill asynchronous PPO from 8 to 64 GPUs, not the thousand-GPU SFT run. A reported task-success improvement used a larger global batch and newer framework; it is not a hardware speedup. No matched foreign time-to-quality, energy or recovery comparison was supplied. Upstream PR checks do not independently reproduce these cluster results.","source_ids":["metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656"]},{"title":"ubenchX core memory-access profiles","evidence":"measured","date":"Reviewed 2026-10-27; publication date unverified","finding":"SemiAnalysis publishes scalar pointer-chase memory profiles for Ascend 950PR_9579 (128 MiB allocation, Cube and Vector controllers) and B200. Values measure differences between matched-address profiles, not direct core-to-core latency or physical distance. Ascend SYS_CNT ticks and B200 clock64 cycles are different units and cannot support a cross-device latency ratio. Concurrent launches lack a grid-wide starting barrier; fine grouping is less repeatable and does not establish physical cache sharing. These are hardware microbenchmarks, not end-to-end AI performance; the Observatory did not rerun them.","source_ids":["inferencex-core-memory-profiles"]},{"title":"DeepInfer published performance scope","evidence":"vendor","date":"2026-10-20","finding":"DeepSeek publishes prefill and decode throughput/latency plots using 4 and 64 NVIDIA SM100 GPUs respectively, each within one scale-up domain. Ascend 950 is a supported software target, but these plots are not Ascend measurements or a matched cross-hardware comparison. The offline release excludes complete service infrastructure. No benchmark was rerun by the Observatory.","source_ids":["deepinfer-release-2026-10-20"]},{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}]
- Updated: CXMT
4 field changes
- assessment: Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. The unaudited nine-month filing falls within the earlier earnings forecast. Sixth-generation DRAM engineering-sample validation is reported, but customer qualification and production remain unverified. Its announced 4F2 VSG roadmap targets a server DDR5 RDIMM by year-end 2026, but no completed product, qualification, capacity, speed or density was disclosed. → Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. The unaudited nine-month filing falls within the earlier earnings forecast. Sixth-generation DRAM engineering-sample validation is reported, but customer qualification and production remain unverified. Its October 30 clarification targets 4F2 DDR5 RDIMM sampling by year-end 2026; volume production remains conditional on customer validation and demand. The earlier October 10 roadmap is retained.
- facts: [{"label":"Sixth-generation DRAM engineering validation","value":"CXMT reports completion of engineering-sample validation for DRAM developed on its sixth-generation platform. Its unaudited nine-month results show RMB 269.024 billion revenue and RMB 132.040 billion attributable net profit, within the earlier forecast. Product specifications, customer qualification, yield and production volume remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["cxmt-q3-2026-10-31"]},{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}] → [{"label":"4F2 RDIMM sampling clarification","value":"CXMT specifies sampling of its first 4F2 DDR5 RDIMM by year-end 2026; volume production depends on customer validation and demand. This is a roadmap target, not completed shipment or qualification.","evidence":"vendor","as_of":"2026-10-30","source_ids":["cxmt-rdimm-sampling-2026-10-31"]},{"label":"Sixth-generation DRAM engineering validation","value":"CXMT reports completion of engineering-sample validation for DRAM developed on its sixth-generation platform. Its unaudited nine-month results show RMB 269.024 billion revenue and RMB 132.040 billion attributable net profit, within the earlier forecast. Product specifications, customer qualification, yield and production volume remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["cxmt-q3-2026-10-31"]},{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}]
- source_ids: ["cxmt-q3-2026-10-31","cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"] → ["cxmt-rdimm-sampling-2026-10-31","cxmt-q3-2026-10-31","cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]
- reviewed_at: 2026-11-01 → 2026-11-02
- Updated: MetaX
4 field changes
- assessment: MetaX reports broadening model and framework coverage and forecast a first-nine-month profit after a year-earlier loss. This strengthens evidence of commercial scale, but the forecast was unaudited and said increased investment in supply-chain companies also contributed to the profit change. Compatibility and financial evidence do not establish application performance, correctness across workloads, memory provenance or frontier parity. → MetaX reports broadening model and framework coverage and forecast a first-nine-month profit after a year-earlier loss. This strengthens evidence of commercial scale, but the forecast was unaudited and said increased investment in supply-chain companies also contributed to the profit change. Compatibility and financial evidence do not establish application performance, correctness across workloads, memory provenance or frontier parity. Its October RLinf disclosure adds vendor training measurements, while the merged upstream PR records narrower hardware tests and outstanding validation of later fixes.
- facts: [{"label":"Nine-month earnings forecast","value":"Company forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, versus a RMB 345.5 million loss a year earlier. The figures were unaudited, and the filing said increased investment in supply-chain companies contributed to the profit turnaround.","evidence":"primary","as_of":"2026-09-30","source_ids":["metax-nine-month-forecast-2026-10-16"]},{"label":"Software compatibility coverage","value":"MetaX reports 37 flagship-model Day-0 adaptations since December 2025 and full support for PyTorch 2.8's 2,410 GPU operators; no reproducible workload benchmark accompanied the claim.","evidence":"vendor","as_of":"2026-09-14","source_ids":["metax-atria-2026-09-14"]}] → [{"label":"RLinf upstream integration","value":"Merged PR 1656 adds MACA support. Author-reported tests used eight C550 GPUs on the original submission; subsequent fixes were not rerun on MetaX hardware. ManiSkill uses CPU physics, and no MetaX CI job was wired in.","evidence":"primary","as_of":"2026-11-02","source_ids":["rlinf-metax-upstream-pr-1656"]},{"label":"Embodied-model training disclosure","value":"MetaX reports Pi0.5/RoboTwin SFT on 1,024 C-series GPUs (128 eight-card nodes), with 20,000 steps completed in 32 hours. These are vendor measurements without matched foreign-hardware time-to-quality or power controls.","evidence":"vendor","as_of":"2026-10-26","source_ids":["metax-rlinf-validation-2026-10-26"]},{"label":"Nine-month earnings forecast","value":"Company forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, versus a RMB 345.5 million loss a year earlier. The figures were unaudited, and the filing said increased investment in supply-chain companies contributed to the profit turnaround.","evidence":"primary","as_of":"2026-09-30","source_ids":["metax-nine-month-forecast-2026-10-16"]},{"label":"Software compatibility coverage","value":"MetaX reports 37 flagship-model Day-0 adaptations since December 2025 and full support for PyTorch 2.8's 2,410 GPU operators; no reproducible workload benchmark accompanied the claim.","evidence":"vendor","as_of":"2026-09-14","source_ids":["metax-atria-2026-09-14"]}]
- source_ids: ["hbm-reuters","metax-atria-2026-09-14","metax-nine-month-forecast-2026-10-16"] → ["metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","hbm-reuters","metax-atria-2026-09-14","metax-nine-month-forecast-2026-10-16"]
- reviewed_at: 2026-10-16 → 2026-11-02
- Added: CXMT specifies year-end 4F2 RDIMM sampling milestone
- Added: MetaX reports RLinf integration and 1,024-GPU supervised fine-tuning
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-11-01 → 2026-11-02
- last_automated_run_at: 2026-11-01 → 2026-11-02
- coverage: October monthly synthesis completed with explicitly partial public-source coverage; company review dates unchanged. See the November 1 daily sweep and monthly run entries for actual scope. → November 2 partial sweep: 12 priority-1 companies and four oldest remaining profiles screened; CXMT and MetaX substantively updated. Search-only profiles retain their actual review dates.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]},{"date":"2026-11-02","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and four oldest remaining profiles with grouped English/Chinese searches overlapping October 25-November 2. Retrieved CXMT October 30 investor-record PDF, resolving the prior full-text access gap and clarifying year-end RDIMM sampling. Read MetaX CNCC recap and followed its RLinf announcement; verified merged upstream PR 1656, original eight-C550 hardware checks and follow-up limitations. Added vendor thousand-GPU SFT evidence with distinct 8-to-64-GPU scaling scope. Only CXMT and MetaX profile dates advanced. Iluvatar October 28 H200-comparison reporting surfaced, but original test report/configurations were not obtained; company news listing rendered no articles. Its headline ratio remains a follow-up lead, not measured benchmark evidence. T-Head search results and developer index did not establish a new reviewed event. Lithography results remained older reporting without verified acceptance/yield. SMIC October 31 notice schedules Q3 results after trading November 12 and a November 13 call; no Q3 results yet verified. Enflame conference claims, optics investor follow-ups and MetaX Q3 filing remain follow-up work. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f, plus InferenceX landing coverage. No new matched contemporary Chinese frontier comparison verified; result artifacts were not exhaustively audited. Coverage remains partial for tenders, private deployments, unindexed disclosures, customer acceptance, yields, unit volumes, power, reliability and component provenance. Historical briefings and forecasts preserved.","source_ids":["cxmt-rdimm-sampling-2026-10-31","metax-rlinf-validation-2026-10-26","rlinf-metax-upstream-pr-1656","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: CXMT October 30 investor call: 4F2 RDIMM sampling target
- Added: MetaX announces RLinf upstream integration and thousand-GPU validation
- Added: RLinf PR 1656: MetaX MACA support and validation limitations
2026-11-01b293be2 ↗
briefing: synthesize October adoption and production dependencies
- Added: Adoption expands while memory and production dependencies constrain scale
- Updated: data/research-state.yaml
3 field changes
- last_automated_run_at: 2026-10-31 → 2026-11-01
- coverage: Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 24-November 1 overlap. Updated NAURA and CXMT from quarterly filings; only their profile dates advanced. Coverage remains partial. → October monthly synthesis completed with explicitly partial public-source coverage; company review dates unchanged. See the November 1 daily sweep and monthly run entries for actual scope.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["amec","biren","cambricon","cxmt","eoptolink","huawei","innolight","jcet","kunlunxin","naura","thead"],"summary":"Prepared the missing October 1-31 monthly briefing using the monthly helper with as-of November 1. Reviewed selected October records, reread consequential sources and ran English/Chinese gap searches. Preserved all earlier editions and dated forecasts. No company profiles were re-dated for this synthesis. Included October NAURA/CXMT filings recorded November 1 on their actual publication dates. Source-access limitations, unverified broker deployment claims and non-exhaustive benchmark coverage are disclosed in the report. Report complete for the closed period; underlying research remains partial.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","deepinfer-release-2026-10-20","spark-audio-domestic-training-2026-10-22","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","cxmt-q3-2026-10-31","amec-q3-2026-10-31","naura-q3-2026-10-30","innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","biren-placement-completion-2026-10-12","jcet-packaging-investment-2026-10-28","semianalysis-ascend-950pr-2026-10-21"]}]
2026-11-01a65918c ↗
research: record NAURA supply constraints and CXMT engineering validation
- Updated: CXMT
4 field changes
- assessment: Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. An unaudited nine-month forecast reports a sharp revenue increase and profit turnaround amid higher DRAM prices and utilization, strengthening commercial-scale evidence without disclosing unit volume, yield or product mix. Its announced 4F2 VSG roadmap targets a server DDR5 RDIMM by year-end 2026, but no completed product, qualification, capacity, speed or density was disclosed. → Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. The unaudited nine-month filing falls within the earlier earnings forecast. Sixth-generation DRAM engineering-sample validation is reported, but customer qualification and production remain unverified. Its announced 4F2 VSG roadmap targets a server DDR5 RDIMM by year-end 2026, but no completed product, qualification, capacity, speed or density was disclosed.
- facts: [{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}] → [{"label":"Sixth-generation DRAM engineering validation","value":"CXMT reports completion of engineering-sample validation for DRAM developed on its sixth-generation platform. Its unaudited nine-month results show RMB 269.024 billion revenue and RMB 132.040 billion attributable net profit, within the earlier forecast. Product specifications, customer qualification, yield and production volume remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["cxmt-q3-2026-10-31"]},{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}]
- source_ids: ["cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"] → ["cxmt-q3-2026-10-31","cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]
- reviewed_at: 2026-10-28 → 2026-11-01
- Updated: NAURA
3 field changes
- facts: [{"label":"Hybrid-bonding acquisition agreement","value":"NAURA Microelectronics signed an agreement to acquire 100% of Nuohe Jianwei for RMB 404.6 million. The target's hybrid-bonding systems were described as being in product validation and market introduction with repeat orders, but closing conditions remained and customer names, accepted production tools, throughput, yield and uptime were not disclosed.","evidence":"vendor","as_of":"2026-10-15","source_ids":["naura-nuohe-jianwei-acquisition-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reported RMB 20.161 billion of first-half revenue, up 24.90% year over year, and RMB 2.677 billion of R&D expense, up 28.87%.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Advanced packaging expansion","value":"Company reports that hybrid-bonding and high-end TSV electroplating tools were rolled out; the filing does not disclose named-customer acceptance, yield or production volume.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Atomic-layer-etch customer use","value":"Company says its ALE tool has achieved customer process use and an undated product disclosure says it entered a well-known customer's production line and completed process validation. The customer, process node, installed units, yield, uptime and production volume were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-primary","naura-ale-2026-09-24"]},{"label":"Near-term delivery constraint","value":"Company says tight delivery of materials from individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-2026-09-24"]}] → [{"label":"Nine-month sales and delivery constraints","value":"NAURA reports RMB 31.467 billion revenue (+15.26%) and RMB 5.591 billion operating cash inflow (+317.84%). It says shortages in certain materials pressure some deliveries, while new-product customer validation and component upgrades contributed to lower margins. Affected inputs and order values are unspecified.","evidence":"vendor","as_of":"2026-09-30","source_ids":["naura-q3-2026-10-30"]},{"label":"Hybrid-bonding acquisition agreement","value":"NAURA Microelectronics signed an agreement to acquire 100% of Nuohe Jianwei for RMB 404.6 million. The target's hybrid-bonding systems were described as being in product validation and market introduction with repeat orders, but closing conditions remained and customer names, accepted production tools, throughput, yield and uptime were not disclosed.","evidence":"vendor","as_of":"2026-10-15","source_ids":["naura-nuohe-jianwei-acquisition-2026-10-16"]},{"label":"First-half commercial scale","value":"Company reported RMB 20.161 billion of first-half revenue, up 24.90% year over year, and RMB 2.677 billion of R&D expense, up 28.87%.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Advanced packaging expansion","value":"Company reports that hybrid-bonding and high-end TSV electroplating tools were rolled out; the filing does not disclose named-customer acceptance, yield or production volume.","evidence":"vendor","as_of":"2026-06-30","source_ids":["naura-2026-h1-report"]},{"label":"Atomic-layer-etch customer use","value":"Company says its ALE tool has achieved customer process use and an undated product disclosure says it entered a well-known customer's production line and completed process validation. The customer, process node, installed units, yield, uptime and production volume were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-primary","naura-ale-2026-09-24"]},{"label":"Near-term delivery constraint","value":"Company says tight delivery of materials from individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed.","evidence":"vendor","as_of":"2026-09-24","source_ids":["naura-ale-2026-09-24"]}]
- source_ids: ["naura","naura-2026-h1-report","naura-ale-primary","naura-ale-2026-09-24","naura-nuohe-jianwei-acquisition-2026-10-16"] → ["naura-q3-2026-10-30","naura","naura-2026-h1-report","naura-ale-primary","naura-ale-2026-09-24","naura-nuohe-jianwei-acquisition-2026-10-16"]
- reviewed_at: 2026-10-16 → 2026-11-01
- Added: CXMT reports sixth-generation DRAM engineering-sample validation
- Added: NAURA quarterly filing reiterates material-supply delivery pressure
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-10-31 → 2026-11-01
- coverage: Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 23-31 overlap. Updated AMEC and Kunlunxin from primary disclosures; only their profile dates advanced. Coverage is partial, with earnings and software follow-ups outstanding. → Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 24-November 1 overlap. Updated NAURA and CXMT from quarterly filings; only their profile dates advanced. Coverage remains partial.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-11-01","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","thead","metax"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles with grouped English and Chinese searches overlapping October 24-November 1. Read NAURA CNInfo quarterly filing and CXMT quarterly filing at Sina, resolving prior earnings follow-ups. Added material-supply delivery constraints and sixth-generation DRAM engineering-sample validation; preserved historical forecasts and earlier production snapshots. Only NAURA and CXMT profile dates advanced. CXMT October 30 investor-call record surfaced with a year-end RDIMM sampling target, but full-page retrieval failed; precise roadmap clarification remains follow-up work. MetaX CNCC recap retrieval timed out; Iluvatar software artifacts and T-Head developments were not substantively verified. Lithography searches returned earlier reports without new verified qualification milestones. Search-only profiles retain dates. Checked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f. Read InferenceX landing coverage; registered Ascend microbenchmarks remain distinct from end-to-end workload comparisons. No new matched contemporary Chinese frontier benchmark was verified; individual artifacts were not exhaustively audited. Coverage remains incomplete for unindexed disclosures, private deployments, tenders, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["naura-q3-2026-10-30","cxmt-q3-2026-10-31","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: CXMT 2026 third-quarter report
- Added: NAURA 2026 third-quarter report
2026-10-31398b1cb ↗
research: add AMEC production scale and Kunlunxin compatibility evidence
- Updated: AMEC
3 field changes
- facts: [{"label":"Reported installed reaction-chamber scale","value":"China Star Market reported that AMEC had more than 8,800 cumulative reaction chambers operating in over 220 domestic and overseas production lines at more than 100 customers by June 2026. The report did not name customers or disclose product mix, nodes, utilization, uptime, yield contribution or repeat-order economics.","evidence":"reported","as_of":"2026-06-30","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Advanced-memory tool launches","value":"AMEC displayed its newly launched Primo XD-RIE 70-to-90:1 aspect-ratio etch system and Prefima Unicore AM 300 mm molybdenum ALD system. Announced capabilities were not accompanied by named-customer qualification, accepted deliveries, production use or operating metrics.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]}] → [{"label":"Reported installed reaction-chamber scale","value":"China Star Market reported that AMEC had more than 8,800 cumulative reaction chambers operating in over 220 domestic and overseas production lines at more than 100 customers by June 2026. The report did not name customers or disclose product mix, nodes, utilization, uptime, yield contribution or repeat-order economics.","evidence":"reported","as_of":"2026-06-30","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Advanced-memory tool launches","value":"AMEC displayed its newly launched Primo XD-RIE 70-to-90:1 aspect-ratio etch system and Prefima Unicore AM 300 mm molybdenum ALD system. Announced capabilities were not accompanied by named-customer qualification, accepted deliveries, production use or operating metrics.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]},{"label":"September production footprint","value":"The unaudited quarterly filing reports more than 9,300 cumulative reaction chambers in volume-production lines domestically and overseas by September 30. Nine-month revenue was RMB 10.326 billion; attributable profit excluding non-recurring items was RMB 1.623 billion. Product mix, named customers, nodes, utilization and yield contribution remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["amec-q3-2026-10-31"]}]
- source_ids: ["amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13","guoxin-expo-2026-10-12"] → ["amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13","guoxin-expo-2026-10-12","amec-q3-2026-10-31"]
- reviewed_at: 2026-10-20 → 2026-10-31
- Updated: Kunlunxin
3 field changes
- facts: [{"label":"Reported China market share","value":"IDC data reported by Reuters put Kunlunxin at 4% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"RLinf distributed-training support","value":"Kunlunxin reports P800 validation across single- and multi-node RLinf modes, with a standard XPU plugin, XCCL collectives and Pi 0.5 LIBERO reproduction; release-v0.4 exposes setup instructions. Reported task and synchronization results lack a matched foreign-hardware control or independent reproduction.","evidence":"primary","as_of":"2026-10-10","source_ids":["kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support"]},{"label":"MiniCPM5-2B deployment","value":"Kunlunxin and FlagOS report Day-0 adaptation and inference validation, with a BF16 deployment image published through ModelScope and Hugging Face.","evidence":"vendor","as_of":"2026-09-10","source_ids":["kunlunxin-minicpm-2026-09-10"]}] → [{"label":"Reported China market share","value":"IDC data reported by Reuters put Kunlunxin at 4% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"RLinf distributed-training support","value":"Kunlunxin reports P800 validation across single- and multi-node RLinf modes, with a standard XPU plugin, XCCL collectives and Pi 0.5 LIBERO reproduction; release-v0.4 exposes setup instructions. Reported task and synchronization results lack a matched foreign-hardware control or independent reproduction.","evidence":"primary","as_of":"2026-10-10","source_ids":["kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support"]},{"label":"MiniCPM5-2B deployment","value":"Kunlunxin and FlagOS report Day-0 adaptation and inference validation, with a BF16 deployment image published through ModelScope and Hugging Face.","evidence":"vendor","as_of":"2026-09-10","source_ids":["kunlunxin-minicpm-2026-09-10"]},{"label":"P800 operating-system compatibility","value":"openEuler reports all seven compatibility subtests passed, covering device identification, driver and SDK deployment, and host/device and device/device links. This is partner-reported runtime compatibility, not independent AI workload performance, reliability at scale or CUDA parity.","evidence":"vendor","as_of":"2026-10-19","source_ids":["kunlunxin-openeuler-2026-10-19"]}]
- source_ids: ["kunlunxin","kunlunxin-minicpm-2026-09-10","china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support"] → ["kunlunxin","kunlunxin-minicpm-2026-09-10","china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kunlunxin-openeuler-2026-10-19"]
- reviewed_at: 2026-10-13 → 2026-10-31
- Added: AMEC reports more than 9,300 reaction chambers in production
- Added: openEuler and Kunlunxin report P800 compatibility certification
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-30 → 2026-10-31
- last_automated_run_at: 2026-10-28 → 2026-10-31
- coverage: Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 22-30 overlap. Added InnoLight, Eoptolink and Kingsemi quarterly filings; only these profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics. → Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 23-31 overlap. Updated AMEC and Kunlunxin from primary disclosures; only their profile dates advanced. Coverage is partial, with earnings and software follow-ups outstanding.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-31","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kunlunxin","thead"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with grouped Chinese and English searches overlapping October 23-31. Read AMEC quarterly filing at the Sina mirror after the SSE PDF was inaccessible; added September cumulative chamber scale while preserving the June snapshot. Read openEuler October 19 announcement and Kunlunxin October 23 repost to resolve a prior follow-up; recorded the earlier publication and left the precise certification event date unknown. Only AMEC and Kunlunxin review dates advanced. NAURA and CXMT quarterly results surfaced but primary filings remain follow-up work. Iluvatar model-adaptation artifacts and T-Head updates were not substantively verified. RSLaser, Aishengna and Yuliangsheng searches mostly repeated earlier DUV reports; no new qualification milestone was established. Search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage; Ascend microbenchmarks are already registered and no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited. Historical reports, corrections and forecasts are preserved. Coverage remains incomplete for private deployments, tenders, unindexed disclosures, customer acceptance, unit shipments, yield, utilization, energy, reliability and component provenance.","source_ids":["amec-q3-2026-10-31","kunlunxin-openeuler-2026-10-19","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: AMEC reports more than 9,300 reaction chambers in production
- Added: openEuler and Kunlunxin report P800 compatibility certification
2026-10-3055cea55 ↗
research: add optical and equipment Q3 operating evidence
- Updated: Eoptolink
3 field changes
- facts: [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"1.6T volume-ramp disclosure","value":"Company says 800G and 1.6T are its main delivered products in 2026 and 1.6T has entered a continuing volume ramp. Faster future shipments remain an expectation; unit counts, yields, utilization and named customer acceptance are undisclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["eoptolink-operations-2026-10-23"]}] → [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"1.6T volume-ramp disclosure","value":"Company says 800G and 1.6T are its main delivered products in 2026 and 1.6T has entered a continuing volume ramp. Faster future shipments remain an expectation; unit counts, yields, utilization and named customer acceptance are undisclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["eoptolink-operations-2026-10-23"]},{"label":"Nine-month growth and cash-flow pressure","value":"The unaudited nine-month filing reports revenue of RMB 31.850 billion, up 92.97%. Operating cash flow fell 68.42% to RMB 1.464 billion, attributed to increased material payments and taxes. September inventory reached RMB 17.413 billion, up 140.70% from year-end, for existing and expected orders and stocking cycles. The filing does not quantify qualified capacity or component provenance.","evidence":"vendor","as_of":"2026-09-30","source_ids":["eoptolink-q3-2026-10-30"]}]
- source_ids: ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22","eoptolink-operations-2026-10-23"] → ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22","eoptolink-operations-2026-10-23","eoptolink-q3-2026-10-30"]
- reviewed_at: 2026-10-24 → 2026-10-30
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]},{"label":"Nine-month sales and inventory expansion","value":"The unaudited nine-month filing reports revenue of RMB 69.547 billion, up 178.13%, with continued 800G/1.6T sales growth. September inventory was RMB 23.878 billion, up 88.30% from year-end, attributed to order-related stockpiling. Operating cash flow was RMB 8.008 billion. These are company financial disclosures, not qualified capacity or component-provenance measurements.","evidence":"vendor","as_of":"2026-09-30","source_ids":["innolight-q3-2026-10-30"]}]
- source_ids: ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23"] → ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23","innolight-q3-2026-10-30"]
- reviewed_at: 2026-10-24 → 2026-10-30
- Updated: Kingsemi
3 field changes
- facts: [{"label":"Reported order and capacity pressure","value":"Management said new orders grew markedly and production was under pressure, prompting capacity expansion; it targets faster front-end track progress in 2026-2027 and further customer introduction of cleaning tools. No order, capacity, accepted-tool or qualification counts were disclosed.","evidence":"reported","as_of":"2026-10-09","source_ids":["kingsemi-capacity-2026-10-10"]},{"label":"First-half product mix","value":"Front-end revenue was reported roughly level with back-end advanced-packaging revenue and cleaning equipment exceeded 30% of revenue; revenue mix does not establish advanced-node qualification.","evidence":"reported","as_of":"2026-06-30","source_ids":["kingsemi-capacity-2026-10-10"]}] → [{"label":"Reported order and capacity pressure","value":"Management said new orders grew markedly and production was under pressure, prompting capacity expansion; it targets faster front-end track progress in 2026-2027 and further customer introduction of cleaning tools. No order, capacity, accepted-tool or qualification counts were disclosed.","evidence":"reported","as_of":"2026-10-09","source_ids":["kingsemi-capacity-2026-10-10"]},{"label":"First-half product mix","value":"Front-end revenue was reported roughly level with back-end advanced-packaging revenue and cleaning equipment exceeded 30% of revenue; revenue mix does not establish advanced-node qualification.","evidence":"reported","as_of":"2026-06-30","source_ids":["kingsemi-capacity-2026-10-10"]},{"label":"Nine-month delivery revenue and operating losses","value":"The unaudited filing reports nine-month revenue of RMB 1.334 billion, up 34.73%, attributable net loss of RMB 55.0 million and operating cash outflow of RMB 446.7 million. Management links revenue growth to orders, deliveries and acceptance; higher personnel and procurement payments reduced cash flow. Accepted-tool counts, process nodes, throughput and defect rates remain undisclosed.","evidence":"vendor","as_of":"2026-09-30","source_ids":["kingsemi-q3-2026-10-30"]}]
- source_ids: ["kingsemi","kingsemi-capacity-2026-10-10"] → ["kingsemi","kingsemi-capacity-2026-10-10","kingsemi-q3-2026-10-30"]
- reviewed_at: 2026-10-13 → 2026-10-30
- Added: Eoptolink reports sales growth with higher inventory and lower operating cash flow
- Added: InnoLight reports optical sales growth alongside inventory expansion
- Added: Kingsemi reports rising delivery revenue but continuing losses and cash outflow
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-10-29 → 2026-10-30
- coverage: Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 21-29 overlap and targeted backfill. Updated Cambricon, Biren and H3C; only these profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics. → Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 22-30 overlap. Added InnoLight, Eoptolink and Kingsemi quarterly filings; only these profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-30","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","iluvatar","kingsemi","kunlunxin"],"summary":"Screened all 12 priority-1 companies plus RSLaser, Iluvatar, Kingsemi and Kunlunxin, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 22-30. Read InnoLight, Eoptolink and Kingsemi unaudited third-quarter filings at Sina mirrors. Formal filing dates are October 30; October 29 press coverage surfaced them earlier. Financial period-end dates remain September 30. Added revenue, inventory and cash-flow evidence without converting asset investment into qualified capacity, stockpiles into shipments, or financial reports into independent technical measurements. Only these three profile dates advanced. Kunlunxin and Iluvatar news indexes surfaced October 23 MiniMax compatibility claims and Kunlunxin/openEuler certification; exact announcement and test artifacts remain follow-up work. RSLaser, Aishengna and Yuliangsheng searches did not establish a new qualification milestone; other search-only profiles retain their dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, unchanged. Read InferenceX landing coverage, whose Ascend microbenchmarks are already registered; no new matched contemporary Chinese frontier end-to-end comparison was verified. Individual benchmark artifacts were not exhaustively audited and no benchmark review date advanced. Historical news, corrections, briefings and dated forecasts are preserved. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yield, utilization, reliability, energy and component provenance.","source_ids":["innolight-q3-2026-10-30","eoptolink-q3-2026-10-30","kingsemi-q3-2026-10-30","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Eoptolink 2026 third-quarter report (unaudited)
- Added: InnoLight 2026 third-quarter report (unaudited)
- Added: Kingsemi 2026 third-quarter report (unaudited)
2026-10-29cf433c7 ↗
research: clarify Cambricon shipments and add Biren and H3C milestones
- Updated: Biren
3 field changes
- facts: [{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]}] → [{"label":"First-half 2026 commercialization","value":"Company reports RMB 1.236 billion revenue (up 1,997.6% year on year) and a RMB 377 million loss for the period in its September 7 half-year summary.","evidence":"vendor","as_of":"2026-06-30","source_ids":["biren-2026-h1"]},{"label":"Model compatibility","value":"Reports functional and accuracy validation of GLM-5.3-Flash on Bili 166M using SGLang and BIRENSUPA; no matched hardware speedup established.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-glm-2026-09-07"]},{"label":"PyTorch backend","value":"Company says it released TorchSUPA, a PrivateUse1 backend plugin with common API, collective-communication, graph-capture and profiling support; no public code repository was identified in this review.","evidence":"vendor","as_of":"2026-09-07","source_ids":["biren-torchsupa-2026-09-07"]},{"label":"Optical-interconnect research","value":"September 2 Shanghai University laboratory agreement includes planned AI-chip optical-interconnect and cooling research, not production qualification.","evidence":"vendor","as_of":"2026-09-02","source_ids":["biren-shanghai-lab-2026-09-10"]},{"label":"Next-generation product validation","value":"Company says an unnamed new-generation product completed bring-up and entered post-silicon validation; customer sampling and volume production remain expected over coming quarters, subject to satisfactory validation and qualification.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"Conditional October placement","value":"Proposed best-efforts placement of 130 million H shares at HK$31.08 each could raise an estimated HK$4.0197 billion net; 70% is allocated to supply-chain procurement, production readiness and commercialization, but the transaction had not completed.","evidence":"vendor","as_of":"2026-10-08","source_ids":["biren-placement-2026-10-08"]},{"label":"October placement completed","value":"The October 12 filing confirms completion of 130 million H shares at HK$31.08, with HK$4.0404 billion gross and approximately HK$4.0197 billion net proceeds. This closes the previously conditional October transaction; funding availability does not establish completed procurement or qualified chip production.","evidence":"primary","as_of":"2026-10-12","source_ids":["biren-placement-completion-2026-10-12"]}]
- source_ids: ["biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08"] → ["biren","biren-2026-h1","biren-glm-2026-09-07","biren-torchsupa-2026-09-07","biren-shanghai-lab-2026-09-10","biren-placement-2026-10-08","biren-placement-completion-2026-10-12"]
- reviewed_at: 2026-10-10 → 2026-10-29
- Updated: Cambricon
3 field changes
- facts: [{"label":"Reported China market share","value":"IDC data reported by Reuters put Cambricon at 5% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Upstream software participation","value":"PyTorch Foundation Platinum member with reported contributions spanning torch.compile, runtime, distributed computing, AMP, data loading and profiling.","evidence":"primary","as_of":"2026-09-08","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09"]},{"label":"Q3 financial scale and working capital","value":"Unaudited Q3 revenue was RMB 3.531 billion and attributable profit RMB 1.404 billion. Operating cash flow was negative RMB 83.68 million for Q3, versus positive RMB 227.64 million for January-September. September-end inventories reached RMB 11.665 billion and prepayments RMB 3.044 billion, versus RMB 4.944 billion and RMB 0.745 billion at year-end 2025. The report does not identify chip/HBM inventory mix, shipment counts or customer utilization.","evidence":"primary","as_of":"2026-09-30","source_ids":["cambricon-q3-2026-report"]}] → [{"label":"Reported China market share","value":"IDC data reported by Reuters put Cambricon at 5% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Upstream software participation","value":"PyTorch Foundation Platinum member with reported contributions spanning torch.compile, runtime, distributed computing, AMP, data loading and profiling.","evidence":"primary","as_of":"2026-09-08","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09"]},{"label":"Q3 financial scale and working capital","value":"Unaudited Q3 revenue was RMB 3.531 billion and attributable profit RMB 1.404 billion. Operating cash flow was negative RMB 83.68 million for Q3, versus positive RMB 227.64 million for January-September. September-end inventories reached RMB 11.665 billion and prepayments RMB 3.044 billion, versus RMB 4.944 billion and RMB 0.745 billion at year-end 2025. The report does not identify chip/HBM inventory mix, shipment counts or customer utilization.","evidence":"primary","as_of":"2026-09-30","source_ids":["cambricon-q3-2026-report"]},{"label":"Shipment guidance and inventory qualification","value":"At the October 26 briefing, management said it had never communicated the million-chip shipment figure. No SiYuan 690 shipment count or timetable was disclosed. Procurement reflects both customer orders and forecast demand; stockpiles therefore do not establish fully contracted shipments. Management reported RMB 501.6129 million of nine-month inventory write-down provisions, mainly associated with aging strategic raw-material stocks.","evidence":"reported","as_of":"2026-10-26","source_ids":["cambricon-shipment-clarification-2026-10-26"]}]
- source_ids: ["hbm-reuters","cambricon","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report"] → ["hbm-reuters","cambricon","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report","cambricon-shipment-clarification-2026-10-26"]
- reviewed_at: 2026-10-26 → 2026-10-29
- Updated: H3C
3 field changes
- facts: [{"label":"Open scale-up direction","value":"H3C describes UXM module support, G-Link development and NPO/CPO interfaces with 112G/224G compatibility and a reserved 448G SerDes path.","evidence":"primary","as_of":"2026-09-04","source_ids":["h3c-odx-2026-09-04"]},{"label":"Domestic-accelerator supernode","value":"H3C says UniPoD S81000 X1 supports 40 unnamed domestic accelerators in one scale-up domain, 448 GB/s peer-to-peer bandwidth, 5,760 GB of unified-address memory and up to 28 PFLOP/s FP8. Accelerator and switch silicon, dense-versus-sparse convention, power, deployment and workload results were not disclosed.","evidence":"vendor","as_of":"2026-09-29","source_ids":["h3c-unipod-s81000-x1-2026-09-29"]}] → [{"label":"Open scale-up direction","value":"H3C describes UXM module support, G-Link development and NPO/CPO interfaces with 112G/224G compatibility and a reserved 448G SerDes path.","evidence":"primary","as_of":"2026-09-04","source_ids":["h3c-odx-2026-09-04"]},{"label":"Domestic-accelerator supernode","value":"H3C says UniPoD S81000 X1 supports 40 unnamed domestic accelerators in one scale-up domain, 448 GB/s peer-to-peer bandwidth, 5,760 GB of unified-address memory and up to 28 PFLOP/s FP8. Accelerator and switch silicon, dense-versus-sparse convention, power, deployment and workload results were not disclosed.","evidence":"vendor","as_of":"2026-09-29","source_ids":["h3c-unipod-s81000-x1-2026-09-29"]},{"label":"NPO interface standardization","value":"H3C reports co-authoring an OPEN NPO electrical-connector specification covering 6.4T/7.2T optical-engine interfaces and a 224G AI Rack interconnect report. It also claims multi-vendor 224G interoperability with 1.6T OSFP modules and DAC cables; no public test harness, error-rate or power results were provided.","evidence":"vendor","as_of":"2026-10-28","source_ids":["h3c-npo-standard-2026-10-28"]},{"label":"Named university deployment","value":"H3C describes a 77-node GPU cluster at Tsinghua Shenzhen International Graduate School using R5300 servers, 200G RoCE, X20000 storage and Kubernetes-based Lingxi management. Its thousand-card framing is not an exact installed-card count; GPU models, domestic origin and independently measured operating results remain undisclosed.","evidence":"vendor","as_of":"2026-10-27","source_ids":["h3c-tsinghua-sig-2026-10-27"]}]
- source_ids: ["h3c","h3c-odx-2026-09-04","h3c-unipod-s81000-x1-2026-09-29"] → ["h3c","h3c-odx-2026-09-04","h3c-unipod-s81000-x1-2026-09-29","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27"]
- reviewed_at: 2026-10-04 → 2026-10-29
- Added: Biren completes October placement with HK$4.0197 billion net proceeds
- Added: Cambricon clarifies shipment rumour and inventory risk
- Added: H3C reports NPO connector standardization and 224G interoperability work
- Added: H3C describes shared 77-node GPU platform at Tsinghua SIG
- Updated: data/research-state.yaml
3 field changes
- last_research_at: 2026-10-28 → 2026-10-29
- coverage: Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with October 20-28 overlap. Added reported CXMT HBM3E plans and JCET investment evidence; only their profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics. → Screened 12 priority-1 companies and four oldest-reviewed remaining profiles with October 21-29 overlap and targeted backfill. Updated Cambricon, Biren and H3C; only these profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-29","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","biren","iluvatar"],"summary":"Screened all 12 priority-1 companies plus RSLaser, H3C, Biren and Iluvatar, the four oldest-reviewed remaining profiles, using overlapping English/Chinese searches for October 21-29. Read original National Business Daily coverage of Cambricon's October 26 briefing; the SSE roadshow opened without text, so its shipment and inventory clarification remains reported evidence. Backfilled Biren's October 12 placement completion from the full HKEX filing, preserving the October 8 conditional announcement. Read H3C's October 27 university-deployment and October 28 NPO announcements; exact GPU count, silicon provenance and independent performance remain unknown. Only Cambricon, Biren and H3C received new profile dates. Other profiles received search screening only, with no new verified qualification milestone for RSLaser, Aishengna or Yuliangsheng; Iluvatar results were largely older disclosures. Optical-policy commentary was not promoted to a new enacted restriction; original regulatory texts and forthcoming earnings remain follow-up work. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified. Individual benchmark artifacts were not exhaustively audited, and no benchmark review date was advanced. Historical news, corrections, briefings and dated forecasts were preserved. Coverage remains incomplete for tenders, private deployments, unindexed disclosures, customer acceptance, yields, shipment volume, utilization, energy, reliability and component provenance.","source_ids":["cambricon-shipment-clarification-2026-10-26","biren-placement-completion-2026-10-12","h3c-npo-standard-2026-10-28","h3c-tsinghua-sig-2026-10-27","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Completion of placing of new H shares and amendments to articles
- Added: Cambricon says it never communicated million-chip shipment figure
- Added: H3C contributes to OPEN NPO connector specification and 224G AI Rack report
- Added: H3C describes Tsinghua SIG shared research-computing platform
2026-10-28532a4df ↗
research: add CXMT HBM3E yield reporting and JCET investment expansion
- Updated: CXMT
3 field changes
- facts: [{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}] → [{"label":"HBM3E limited-production plan and yield estimates","value":"Reuters sources report HBM3E samples sent to domestic chip designers, with limited production expected within two months and estimated yields of 20-30%. CXMT did not comment; qualified shipment volume remains unverified.","evidence":"reported","as_of":"2026-10-28","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]},{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}]
- source_ids: ["cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11"] → ["cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11","cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay"]
- reviewed_at: 2026-10-20 → 2026-10-28
- Updated: JCET
3 field changes
- facts: [{"label":"Computing-electronics mix","value":"First-half computing-electronics revenue grew 40.4% year over year and represented 30.1% of company revenue.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Silicon-photonics validation","value":"Company reports delivering XDFOI-based silicon-photonics engine samples that passed customer validation and working with multiple customers on packaging, thermal and reliability optimization; volume and capacity were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Planned packaging-capacity financing","value":"JCET proposes raising up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion and RMB 1.0 billion for a high-density, high-capacity memory-chip system-level packaging and test upgrade. The offering and projects remain subject to approvals and completion.","evidence":"primary","as_of":"2026-09-03","source_ids":["jcet-private-placement-plan-2026-09-04"]}] → [{"label":"Advanced-packaging investment expansion","value":"At its October 27 briefing, JCET said nine-month fixed capital expenditure was nearly RMB 8.9 billion. Its 2026 investment plan rises from RMB 11.01 billion to RMB 18.2 billion, including fixed-asset investment of RMB 17.31 billion; incremental spending mainly targets advanced-packaging lines.","evidence":"reported","as_of":"2026-10-27","source_ids":["jcet-packaging-investment-2026-10-28"]},{"label":"Computing-electronics mix","value":"First-half computing-electronics revenue grew 40.4% year over year and represented 30.1% of company revenue.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Silicon-photonics validation","value":"Company reports delivering XDFOI-based silicon-photonics engine samples that passed customer validation and working with multiple customers on packaging, thermal and reliability optimization; volume and capacity were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["jcet-2026-h1-report"]},{"label":"Planned packaging-capacity financing","value":"JCET proposes raising up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion and RMB 1.0 billion for a high-density, high-capacity memory-chip system-level packaging and test upgrade. The offering and projects remain subject to approvals and completion.","evidence":"primary","as_of":"2026-09-03","source_ids":["jcet-private-placement-plan-2026-09-04"]}]
- source_ids: ["jcet","jcet-2026-h1-report","jcet-private-placement-plan-2026-09-04"] → ["jcet","jcet-2026-h1-report","jcet-private-placement-plan-2026-09-04","jcet-packaging-investment-2026-10-28"]
- reviewed_at: 2026-10-07 → 2026-10-28
- Added: CXMT reportedly targets limited HBM3E production with low yields
- Added: JCET raises investment plan with advanced-packaging focus
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-27 → 2026-10-28
- last_automated_run_at: 2026-10-27 → 2026-10-28
- coverage: Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with October 19-27 overlap. Backfilled two primary filings and added an independent memory-microbenchmark study. Only Zhiqi and Hwatsing profile dates advanced. Coverage remains partial; search-only checks do not establish complete review of unindexed disclosures, private deployments or production metrics. → Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with October 20-28 overlap. Added reported CXMT HBM3E plans and JCET investment evidence; only their profile dates advanced. Coverage remains partial for unindexed disclosures, private deployments and production metrics.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-28","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","h3c","jcet","biren"],"summary":"Screened the 12 priority-1 companies plus RSLaser, H3C, JCET and Biren, the four oldest-reviewed remaining profiles, with overlapping English/Chinese searches from October 20. Added Reuters reporting on CXMT HBM3E samples, limited-production plans and low yield estimates; original-page retrieval failed, while the search service returned the Reuters syndication text. Syndication is not independent corroboration. Preserved the earlier August small-volume report and all forecasts. Read Securities Daily reporting of JCET's October 27 briefing; the SSE roadshow returned indexed management answers but an empty opened page, so the new facts retain reported evidence labels. Only CXMT and JCET profile dates advanced. H3C searches repeated September announcements; RSLaser, Aishengna and Yuliangsheng produced no verified new qualification milestone. Biren search did not establish placement completion. Cambricon October 26 briefing coverage surfaced as a follow-up for shipment-rumour and supply-chain clarification; it was not promoted to a verified new fact. Remaining search-only profiles keep their actual review dates. Checked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 head 4bb63cd28eb136be221bc801a9e38c66c8a7553f through GitHub, both unchanged. Read InferenceX landing coverage; no new matched contemporary Chinese frontier end-to-end result was verified, and no exhaustive artifact audit was performed. No historical news, correction, briefing or dated forecast was changed. Private deployments, tenders, customer acceptance, shipment volume, yield verification, utilization, energy, reliability and component provenance remain incompletely covered.","source_ids":["cxmt-hbm3e-yields-2026-10-28","cxmt-hbm3e-yields-2026-10-28-relay","jcet-packaging-investment-2026-10-28","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Reuters relay: CXMT HBM3E production plans and yields
- Added: CXMT plans limited HBM3E production despite low yields
- Added: JCET adds advanced-packaging investment
2026-10-27220d9da ↗
research: backfill optics and financing evidence; qualify memory microbenchmarks
- Updated: data/benchmarks.yaml
3 field changes
- assessment: Follow-up through October 26 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 27 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-26 → 2026-10-27
- studies: [{"title":"DeepInfer published performance scope","evidence":"vendor","date":"2026-10-20","finding":"DeepSeek publishes prefill and decode throughput/latency plots using 4 and 64 NVIDIA SM100 GPUs respectively, each within one scale-up domain. Ascend 950 is a supported software target, but these plots are not Ascend measurements or a matched cross-hardware comparison. The offline release excludes complete service infrastructure. No benchmark was rerun by the Observatory.","source_ids":["deepinfer-release-2026-10-20"]},{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}] → [{"title":"ubenchX core memory-access profiles","evidence":"measured","date":"Reviewed 2026-10-27; publication date unverified","finding":"SemiAnalysis publishes scalar pointer-chase memory profiles for Ascend 950PR_9579 (128 MiB allocation, Cube and Vector controllers) and B200. Values measure differences between matched-address profiles, not direct core-to-core latency or physical distance. Ascend SYS_CNT ticks and B200 clock64 cycles are different units and cannot support a cross-device latency ratio. Concurrent launches lack a grid-wide starting barrier; fine grouping is less repeatable and does not establish physical cache sharing. These are hardware microbenchmarks, not end-to-end AI performance; the Observatory did not rerun them.","source_ids":["inferencex-core-memory-profiles"]},{"title":"DeepInfer published performance scope","evidence":"vendor","date":"2026-10-20","finding":"DeepSeek publishes prefill and decode throughput/latency plots using 4 and 64 NVIDIA SM100 GPUs respectively, each within one scale-up domain. Ascend 950 is a supported software target, but these plots are not Ascend measurements or a matched cross-hardware comparison. The offline release excludes complete service infrastructure. No benchmark was rerun by the Observatory.","source_ids":["deepinfer-release-2026-10-20"]},{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}]
- Updated: Hwatsing
3 field changes
- facts: [{"label":"Equipment and capacity financing","value":"Hwatsing completed a private placement of 16,080,508 shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D; schedules, installed capacity and realized output remain future work.","evidence":"primary","as_of":"2026-09-30","source_ids":["hwatsing-placement-2026-09-30"]},{"label":"Process scope","value":"The vendor describes automated endpoint detection and multiple polishing modes for large-format panel-level packaging; customer and production metrics were not disclosed.","evidence":"reported","as_of":"2026-09-09","source_ids":["hwatsing-panel-cmp-2026-09-09"]}] → [{"label":"Placement share registration","value":"Hwatsing reports completing registration of the placement's 16,080,508 new shares on October 15, with net proceeds of RMB 3.7731166749 billion. This is the same financing already recorded, not additional funding or evidence of completed production capacity.","evidence":"primary","as_of":"2026-10-15","source_ids":["hwatsing-share-registration-2026-10-17"]},{"label":"Equipment and capacity financing","value":"Hwatsing completed a private placement of 16,080,508 shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D; schedules, installed capacity and realized output remain future work.","evidence":"primary","as_of":"2026-09-30","source_ids":["hwatsing-placement-2026-09-30"]},{"label":"Process scope","value":"The vendor describes automated endpoint detection and multiple polishing modes for large-format panel-level packaging; customer and production metrics were not disclosed.","evidence":"reported","as_of":"2026-09-09","source_ids":["hwatsing-panel-cmp-2026-09-09"]}]
- source_ids: ["hwatsing","hwatsing-panel-cmp-2026-09-09","hwatsing-placement-2026-09-30"] → ["hwatsing","hwatsing-panel-cmp-2026-09-09","hwatsing-placement-2026-09-30","hwatsing-share-registration-2026-10-17"]
- reviewed_at: 2026-10-05 → 2026-10-27
- Updated: Fujian Zhiqi Photonics
4 field changes
- assessment: The parent company reports growth at Zhiqi, but does not disclose projection-lens qualification, optical precision, yield or scanner deployment. A development program is not a proven EUV optics substitute. → The parent company reports commercial optical-component revenue at Zhiqi, but does not disclose projection-lens qualification, optical precision, yield or scanner deployment. A development program is not a proven EUV optics substitute.
- facts: [{"label":"Parent-company operating update","value":"Fujing Technology said Zhiqi grew quickly in the first half of 2026 and expected full-year development to remain on plan; it did not disclose revenue, customers, lithography qualification, optics specifications, yield or production volume.","evidence":"primary","as_of":"2026-09-01","source_ids":["zhiqi-fujing-ir-2026-09-02"]}] → [{"label":"2025 optical-component operations","value":"Fujing reported Zhiqi revenue of RMB 167 million, up 119.81%, and net profit of RMB 40.0586 million for 2025. Its products include precision lenses, prisms and super-smooth mirrors; this does not establish scanner projection-lens qualification.","evidence":"primary","as_of":"2025-12-31","source_ids":["zhiqi-fujing-ir-2026-05-29"]},{"label":"Parent-company operating update","value":"Fujing Technology said Zhiqi grew quickly in the first half of 2026 and expected full-year development to remain on plan; it did not disclose revenue, customers, lithography qualification, optics specifications, yield or production volume.","evidence":"primary","as_of":"2026-09-01","source_ids":["zhiqi-fujing-ir-2026-09-02"]}]
- source_ids: ["duv-ft","zhiqi-fujing-ir-2026-09-02"] → ["duv-ft","zhiqi-fujing-ir-2026-09-02","zhiqi-fujing-ir-2026-05-29"]
- reviewed_at: 2026-10-02 → 2026-10-27
- Added: Hwatsing completes registration of private-placement shares
- Added: Fujing reports Zhiqi 2025 optical-component revenue and profit
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-26 → 2026-10-27
- last_automated_run_at: 2026-10-26 → 2026-10-27
- coverage: Completed the October 19-25 weekly synthesis with eight in-period events and supplemental source checks. Added one developer-reported training event; preserved all prior editions, company review dates and forecasts. Research coverage remains partial for unindexed disclosures, private operations and unmatched benchmark evidence. → Screened 12 priority-1 companies and the four oldest-reviewed remaining profiles with October 19-27 overlap. Backfilled two primary filings and added an independent memory-microbenchmark study. Only Zhiqi and Hwatsing profile dates advanced. Coverage remains partial; search-only checks do not establish complete review of unindexed disclosures, private deployments or production metrics.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]},{"date":"2026-10-27","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","h3c","hwatsing"],"summary":"Screened 12 priority-1 companies and the four oldest-reviewed non-priority-1 profiles using overlapping English and Chinese searches from October 19, seven days before the preceding run. Most results were older or already recorded. Read Fujing investor-relations record 2026002 and verified its May 29 publication date at the Sina filing mirror; backfilled Zhiqi 2025 operations without treating component sales as scanner qualification. Read Hwatsing announcement 2026-073, recording October 15 registration of the existing placement and net proceeds, not a second raise. Only Zhiqi and Hwatsing profile dates advanced. H3C homepages were accessible but supplied no new qualifying milestone; Hwatsing homepage timed out. RSLaser, Aishengna and Yuliangsheng searches did not establish new primary qualification evidence. Other search-only profiles retain their review dates. Read InferenceX landing page and ubenchX core-memory-profile methodology; added independent hardware measurements with device-specific-unit and repeatability caveats, without calculating an AI speedup. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; individual result artifacts were not exhaustively audited. No new matched current frontier end-to-end comparison was verified. Historical reports, corrections and dated forecasts are unchanged. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, yield, shipment volume, utilization, reliability, power and component provenance.","source_ids":["zhiqi-fujing-ir-2026-05-29","hwatsing-share-registration-2026-10-17","inferencex-core-memory-profiles","h3c","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Hwatsing private-placement results and share-capital change announcement 2026-073
- Added: ubenchX core memory-access profiles: Ascend 950PR and B200
- Added: Fujing Technology investor-relations record 2026002: Zhiqi 2025 operations
2026-10-265f2a799 ↗
briefing: synthesize October 19–25 software and operating evidence
- Added: Software and operating evidence deepen; frontier efficiency remains unproven
- Added: iFlytek reports domestic-hardware training of Spark-Audio
- Updated: data/research-state.yaml
2 field changes
- coverage: Screened 12 priority-1 companies plus oldest-reviewed RSLaser, Zhiqi, ACM and Empyrean over October 18-26, with older filing backfills. Added three events and substantive updates to Cambricon, ACM and Empyrean; only those profile review dates advanced. Coverage remains partial for sparse search results and unresolved primary-source checks. → Completed the October 19-25 weekly synthesis with eight in-period events and supplemental source checks. Added one developer-reported training event; preserved all prior editions, company review dates and forecasts. Research coverage remains partial for unindexed disclosures, private operations and unmatched benchmark evidence.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["huawei","cambricon","innolight","eoptolink","acm"],"summary":"Published the closed October 19-25 weekly synthesis from eight in-period events. Re-read primary filings, DeepInfer release documentation, SemiAnalysis measurements and the accessible training-plan relay. Supplemental English/Chinese cross-sector searches found the October 22 Spark-Audio developer report, now recorded with vendor attribution. Cambricon and ACM were already backfilled by the earlier daily run. No full-profile sweep or company review-date advancement is claimed. Rechecked MLPerf Training and InferenceX web coverage; MLPerf Inference web retrieval failed and its status relies on the earlier same-day GitHub check. No exhaustive artifact audit or new matched frontier comparison. Preserved all historical briefings, corrections and dated forecasts. Unindexed disclosures, private deployments, acceptance, yield, utilization, energy and provenance remain gaps.","source_ids":["acm-hot-spm-recycling-2026-10-21","cambricon-q3-2026-report","deepinfer-release-2026-10-20","deepseek-huawei-design-information-2026-10-19","deepseek-huawei-design-investing-2026-10-19","eoptolink-operations-2026-10-23","inferencex","innolight-operations-2026-10-23","mlperf-inference-6-1","mlperf-training","optical-operations-cnr-2026-10-22","semianalysis-ascend-950pr-2026-10-21","spark-audio-domestic-training-2026-10-22"]}]
- Added: Spark-Audio 1.0 Technical Report (v1)
2026-10-26a9eecb8 ↗
research: add Cambricon quarterly filing and equipment/EDA evidence
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 25 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 26 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-25 → 2026-10-26
- Updated: ACM Research Shanghai
3 field changes
- facts: [] → [{"label":"Parent-reported hot SPM validation","value":"US-listed parent ACM Research says a foundry validated its 170 C single-wafer hot SPM system across more than 50 process layers. It claims 70% sulfuric-acid recycling and expects roughly 400 tons saved per tool annually. Customer identity, node, tool count, wafer volume, yield and independently verified savings were not disclosed.","evidence":"vendor","as_of":"2026-10-21","source_ids":["acm-hot-spm-recycling-2026-10-21"]}]
- source_ids: ["acm","acm-orders-2026-10-01"] → ["acm","acm-orders-2026-10-01","acm-hot-spm-recycling-2026-10-21"]
- reviewed_at: 2026-10-03 → 2026-10-26
- Updated: Cambricon
4 field changes
- assessment: IDC-reported first-half share and upstream PyTorch participation strengthen adoption and ecosystem evidence, but do not identify product mix, customer utilization or matched frontier-workload performance. → Revenue growth supports commercialization, while inventories and prepayments increased sharply. Financial scale and upstream software participation do not establish customer utilization, replenishable HBM supply or matched frontier-workload performance.
- facts: [{"label":"Reported China market share","value":"IDC data reported by Reuters put Cambricon at 5% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Upstream software participation","value":"PyTorch Foundation Platinum member with reported contributions spanning torch.compile, runtime, distributed computing, AMP, data loading and profiling.","evidence":"primary","as_of":"2026-09-08","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09"]}] → [{"label":"Reported China market share","value":"IDC data reported by Reuters put Cambricon at 5% of China's data-center AI-accelerator shipments in the first half of 2026; unit volume, product mix, customers and workloads were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Upstream software participation","value":"PyTorch Foundation Platinum member with reported contributions spanning torch.compile, runtime, distributed computing, AMP, data loading and profiling.","evidence":"primary","as_of":"2026-09-08","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09"]},{"label":"Q3 financial scale and working capital","value":"Unaudited Q3 revenue was RMB 3.531 billion and attributable profit RMB 1.404 billion. Operating cash flow was negative RMB 83.68 million for Q3, versus positive RMB 227.64 million for January-September. September-end inventories reached RMB 11.665 billion and prepayments RMB 3.044 billion, versus RMB 4.944 billion and RMB 0.745 billion at year-end 2025. The report does not identify chip/HBM inventory mix, shipment counts or customer utilization.","evidence":"primary","as_of":"2026-09-30","source_ids":["cambricon-q3-2026-report"]}]
- source_ids: ["hbm-reuters","cambricon","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","china-ai-accelerator-share-reuters-2026-10-13"] → ["hbm-reuters","cambricon","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","china-ai-accelerator-share-reuters-2026-10-13","cambricon-q3-2026-report"]
- reviewed_at: 2026-10-13 → 2026-10-26
- Updated: Empyrean
3 field changes
- facts: [{"label":"Customer-flow and foundry-certification claims","value":"Empyrean says products entered core design flows at leading domestic chip-design and manufacturing customers and supported mass production. It reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm; the filing does not name foundries, customers, tape-outs, covered rule decks, sign-off correlation or production volume.","evidence":"primary","as_of":"2026-06-30","source_ids":["empyrean-2026-h1-report"]}] → [{"label":"Customer-flow and foundry-certification claims","value":"Empyrean says products entered core design flows at leading domestic chip-design and manufacturing customers and supported mass production. It reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm; the filing does not name foundries, customers, tape-outs, covered rule decks, sign-off correlation or production volume.","evidence":"primary","as_of":"2026-06-30","source_ids":["empyrean-2026-h1-report"]},{"label":"Proposed hardware-emulation investment","value":"On October 13 the board approved contributing RMB 150 million cash and its indirectly held 7.78% S2C stake to Wuxi Yake Hongyu as part of a proposed restructuring. Cash funding is conditional on the merger; the filing says the core hardware-emulation accelerator remains in R&D, with validation and commercialization uncertain. It does not establish transaction completion or a qualified complete digital flow.","evidence":"primary","as_of":"2026-10-13","source_ids":["empyrean-yake-s2c-investment-2026-10-13"]}]
- source_ids: ["empyrean","empyrean-2026-h1-report"] → ["empyrean","empyrean-2026-h1-report","empyrean-yake-s2c-investment-2026-10-13"]
- reviewed_at: 2026-10-03 → 2026-10-26
- Added: ACM reports foundry validation of hot SPM cleaning and acid recycling
- Added: Cambricon reports Q3 growth alongside larger inventory and prepayments
- Added: Empyrean proposes capital and S2C stake contribution for hardware emulation
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-25 → 2026-10-26
- last_automated_run_at: 2026-10-25 → 2026-10-26
- coverage: Screened 12 priority-1 companies plus oldest-reviewed RSLaser, SMEE, Zhiqi and CIOMP over October 17-25. No new material fact was verified. Reviewed SMEE portfolio/news and CIOMP Chinese research listings; only SMEE and CIOMP profile review dates advanced. Coverage remains partial because results were often stale or sparse and several primary follow-ups remain unresolved. → Screened 12 priority-1 companies plus oldest-reviewed RSLaser, Zhiqi, ACM and Empyrean over October 18-26, with older filing backfills. Added three events and substantive updates to Cambricon, ACM and Empyrean; only those profile review dates advanced. Coverage remains partial for sparse search results and unresolved primary-source checks.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-26","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","zhiqi","acm","empyrean"],"summary":"Searched English and Chinese names over October 18-26, starting seven days before the preceding run. Read Cambricon Q3 filing and verified its unaudited revenue, cash-flow, inventory and prepayment figures; financial values do not establish shipped units or HBM provenance. Read ACM parent October 21 hot-SPM announcement and backfilled Empyrean October 13 investment filing discovered through October 20 merger reporting. Kept customer-validation claims attributed and the restructuring conditional, with hardware emulation still in R&D. Empyrean English portfolio/news pages were read; its Chinese site failed to load. ACM Shanghai homepage timed out, so parent evidence is explicitly identified. RSLaser searches mainly found patents and recycled investment stories; Zhiqi results included recruitment, not scanner qualification. Other priority-company searches were often stale, sparse or repetitions of recorded events. AMEC order commentary syndicated on October 22 traces to earlier Bay Area exhibition reporting with inconsistent displayed dates; technical claims await primary confirmation and no new fact was added. Previous NAURA award and SMEE SSD820 follow-ups remain unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked via GitHub. Read the InferenceX landing page; no new matched frontier end-to-end comparison was verified. Existing independent Ascend kernel evidence is retained. Individual benchmark artifacts were not exhaustively audited. No historical briefing, correction or dated forecast changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, shipment volume, yields, utilization, reliability, power and component provenance.","source_ids":["cambricon-q3-2026-report","acm-hot-spm-recycling-2026-10-21","empyrean-yake-s2c-investment-2026-10-13","empyrean","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: ACM Research enhances single-wafer hot SPM system with sulfuric acid recycle capability
- Added: Cambricon 2026 third-quarter report (board dated October 23; disclosure dated October 24)
- Added: Empyrean announcement 2026-049: proposed investment in associate and related-party transaction
- Updated: EDA portfolio
1 field changes
- reviewed_at: 2026-09-13 → 2026-10-26
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-25 → 2026-10-26
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-25 → 2026-10-26
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-25 → 2026-10-26
2026-10-25a73a9ed ↗
research: record October 25 overlap review and unresolved follow-ups
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 24 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 25 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-24 → 2026-10-25
- Updated: CIOMP
2 field changes
- source_ids: ["ciomp","euv-reuters"] → ["ciomp","euv-reuters","ciomp-chinese-research"]
- reviewed_at: 2026-10-03 → 2026-10-25
- Updated: SMEE
1 field changes
- reviewed_at: 2026-09-18 → 2026-10-25
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-24 → 2026-10-25
- last_automated_run_at: 2026-10-24 → 2026-10-25
- coverage: Screened all 12 priority-1 companies and the four oldest-reviewed remaining profiles (RSLaser, SMEE, SIOM and Zhiqi) over October 16-24. Read original optical operating filings and SIOM research disclosure; only InnoLight, Eoptolink and SIOM profile dates advanced. Sparse/stale search results, private deployments, tenders and undisclosed production metrics remain coverage gaps. → Screened 12 priority-1 companies plus oldest-reviewed RSLaser, SMEE, Zhiqi and CIOMP over October 17-25. No new material fact was verified. Reviewed SMEE portfolio/news and CIOMP Chinese research listings; only SMEE and CIOMP profile review dates advanced. Coverage remains partial because results were often stale or sparse and several primary follow-ups remain unresolved.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-25","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","zhiqi","ciomp"],"summary":"Screened priority and oldest-reviewed profiles with overlapping English and Chinese searches from October 17, seven days before the preceding run. Most results repeated existing evidence or were stale. Read NAURA homepage, which confirms the October 13 supplier-award headline, but its linked WeChat detail could not be retrieved and a targeted Infineon search did not establish tool, fab or qualification details; retained the follow-up without adding a production claim. Read SMEE portfolio and company-news listings; no SSD820 primary release was located. Exhibition reporting remains a follow-up, not verified production acceptance. Read CIOMP Chinese homepage and research listings; no new qualifying integrated scanner or projection-optics milestone was established. Its English site timed out. SMEE and CIOMP received actual review dates without factual changes; search-only profiles retained their dates. RSLaser and Zhiqi search results did not establish new qualified deliveries. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Read InferenceX current landing page; no new matched frontier end-to-end comparison was verified, and existing independent Ascend kernel evidence remains intact. Individual result artifacts were not exhaustively audited. No news, forecast, historical briefing or correction changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, shipment quantities, yield, customer acceptance, utilization, reliability, power and component provenance.","source_ids":["naura","smee","ciomp-chinese-research","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: CIOMP Chinese research and news listings
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-24 → 2026-10-25
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-24 → 2026-10-25
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-24 → 2026-10-25
- Updated: Lithography equipment portfolio
1 field changes
- reviewed_at: 2026-09-18 → 2026-10-25
2026-10-240094a65 ↗
research: add optical order and shipment disclosures for October 24
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 23 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 24 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-23 → 2026-10-24
- Updated: Eoptolink
3 field changes
- facts: [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]}] → [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"1.6T volume-ramp disclosure","value":"Company says 800G and 1.6T are its main delivered products in 2026 and 1.6T has entered a continuing volume ramp. Faster future shipments remain an expectation; unit counts, yields, utilization and named customer acceptance are undisclosed.","evidence":"vendor","as_of":"2026-10-23","source_ids":["eoptolink-operations-2026-10-23"]}]
- source_ids: ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22"] → ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22","eoptolink-operations-2026-10-23"]
- reviewed_at: 2026-10-23 → 2026-10-24
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]},{"label":"Order horizon and next-generation sampling","value":"Company says major overseas customers have placed orders through 2027, while some key customers supplied 2028 demand guidance. It reports co-development and sample delivery for unspecified next-generation products; neither guidance nor samples establish firm 2028 orders or qualified volume production.","evidence":"vendor","as_of":"2026-10-23","source_ids":["innolight-operations-2026-10-23"]}]
- source_ids: ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22"] → ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22","innolight-operations-2026-10-23"]
- reviewed_at: 2026-10-23 → 2026-10-24
- Updated: SIOM
3 field changes
- facts: [] → [{"label":"Active-glass photonics research","value":"The institute reports molecular sol-gel design combined with DLP 3D printing to make rare-earth-doped active glass and demonstrate optical amplification in a glass waveguide. This is laboratory materials and device research, not evidence of a qualified lithography source or commercial AI-network deployment.","evidence":"primary","as_of":"2026-10-23","source_ids":["siom-active-glass-2026-10-23"]}]
- source_ids: ["siom"] → ["siom","siom-active-glass-2026-10-23"]
- reviewed_at: 2026-10-02 → 2026-10-24
- Added: Optical suppliers disclose order horizon and 1.6T shipment ramp
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-23 → 2026-10-24
- last_automated_run_at: 2026-10-23 → 2026-10-24
- coverage: Screened all 12 priority-1 companies plus the four oldest-reviewed remaining profiles (RSLaser, Guowang, SMEE and RoboTechnik), using overlapping English and Chinese searches for October 15-23 and targeted primary-source backfills. Reviewed supporting documents for Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik; only those five profile dates advanced. Coverage remains partial: several searches returned stale or sparse results, and private deployments, tenders, production metrics and individual benchmark artifacts were not comprehensively audited. → Screened all 12 priority-1 companies and the four oldest-reviewed remaining profiles (RSLaser, SMEE, SIOM and Zhiqi) over October 16-24. Read original optical operating filings and SIOM research disclosure; only InnoLight, Eoptolink and SIOM profile dates advanced. Sparse/stale search results, private deployments, tenders and undisclosed production metrics remain coverage gaps.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-24","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","smee","siom","zhiqi"],"summary":"Screened 16 companies using overlapping English and Chinese searches from October 16, seven days before the preceding run. Read October 23 InnoLight and Eoptolink original filing PDFs via Sina after the HKEX mirror could not be retrieved. Added order-horizon, sampling and shipment-ramp claims with vendor attribution; primary sourcing is now established, but quantitative optical shipment evidence remains missing. Read SIOM active-glass research announcement and recorded laboratory-stage evidence without advancing commercial qualification. Only these three profile dates advanced. RSLaser and Zhiqi searches yielded no verified new qualifying milestone; an SMEE SSD820 exhibition claim in a user-posted article remains unverified and requires primary follow-up. NAURA supplier-award detail remains unresolved. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub. Reviewed the current InferenceX landing page; no new matched frontier end-to-end result was verified. Existing independent Ascend kernel evidence is retained; individual result artifacts were not exhaustively audited. No dated forecast, historical news, correction or briefing changed. Search-only profiles retain previous review dates. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incompletely covered.","source_ids":["innolight-operations-2026-10-23","eoptolink-operations-2026-10-23","siom-active-glass-2026-10-23","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Eoptolink: voluntary production and operations disclosure (2026-061)
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-22 → 2026-10-24
- Added: InnoLight: voluntary production and operations disclosure (2026-100)
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-22 → 2026-10-24
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-22 → 2026-10-24
- Added: SIOM reports 3D-printed rare-earth active glass for integrated photonics
2026-10-23c2cd7ac ↗
research: add DeepInfer support and optical-sector evidence
- Updated: data/benchmarks.yaml
3 field changes
- assessment: Follow-up through October 22 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 23 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-22 → 2026-10-23
- studies: [{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}] → [{"title":"DeepInfer published performance scope","evidence":"vendor","date":"2026-10-20","finding":"DeepSeek publishes prefill and decode throughput/latency plots using 4 and 64 NVIDIA SM100 GPUs respectively, each within one scale-up domain. Ascend 950 is a supported software target, but these plots are not Ascend measurements or a matched cross-hardware comparison. The offline release excludes complete service infrastructure. No benchmark was rerun by the Observatory.","source_ids":["deepinfer-release-2026-10-20"]},{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}]
- Updated: Eoptolink
3 field changes
- facts: [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]}] → [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]},{"label":"October operating statement","value":"Eoptolink told CNR on October 22 that production and operations were orderly and order deliveries continued normally. This qualitative company statement provides no new shipment volume, yield or product-specific customer acceptance measurements.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]}]
- source_ids: ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11"] → ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11","optical-operations-cnr-2026-10-22"]
- reviewed_at: 2026-10-22 → 2026-10-23
- Updated: Beijing Guowang
3 field changes
- facts: [] → [{"label":"Optical manufacturing recruitment","value":"An October 16 Hebei University notice recruits 2027 graduates for optical coating, processing and inspection, optical-mechanical integration, and system testing. It supports continuing staffing activity but identifies no qualified projection-optics product, delivered unit, customer acceptance or production yield.","evidence":"primary","as_of":"2026-10-16","source_ids":["guowang-recruitment-2026-10-16"]}]
- source_ids: ["guowang"] → ["guowang","guowang-recruitment-2026-10-16"]
- reviewed_at: 2026-09-14 → 2026-10-23
- Updated: Huawei
3 field changes
- facts: [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]},{"label":"Independent Ascend 950PR kernel measurements","value":"SemiAnalysis tested a 950PR_9579 with 28 Cube cores and 128 GiB HBM. These measurements do not establish end-to-end model parity.","evidence":"measured","as_of":"2026-10-21","source_ids":["semianalysis-ascend-950pr-2026-10-21"]}] → [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]},{"label":"Independent Ascend 950PR kernel measurements","value":"SemiAnalysis tested a 950PR_9579 with 28 Cube cores and 128 GiB HBM. These measurements do not establish end-to-end model parity.","evidence":"measured","as_of":"2026-10-21","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"label":"DeepInfer public inference engine","value":"DeepSeek released an Apache-2.0 offline engine for V4.1-Flash supporting Ascend 950 with CANN 9.2 and NVIDIA SM100. The README excludes the KV-cache service, complete disaggregation and replica load balancing; batch-invariant inference and deterministic sampling are NVIDIA-only. Its published performance plots use NVIDIA hardware, not Ascend. Code availability does not establish measured production parity.","evidence":"primary","as_of":"2026-10-20","source_ids":["deepinfer-release-2026-10-20"]}]
- source_ids: ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19","semianalysis-ascend-950pr-2026-10-21"] → ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19","semianalysis-ascend-950pr-2026-10-21","deepinfer-release-2026-10-20"]
- reviewed_at: 2026-10-22 → 2026-10-23
- Updated: InnoLight
3 field changes
- facts: [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]}] → [{"label":"Hong Kong financing","value":"The July 2026 global offering priced at HKD 980 per share with estimated net proceeds of HKD 52.891 billion before any over-allotment exercise.","evidence":"primary","as_of":"2026-07-29","source_ids":["innolight-hk-listing-2026-07-29"]},{"label":"1.6T commercialization","value":"Company says silicon-photonics modules are scaling with shipments rising each quarter, according to National Business Daily's September 11 reporting.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered InnoLight optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"October operating statement","value":"CNR reports that InnoLight said operations and its supply chain were normal, capacity utilization was adequate and it could obtain and deliver orders. No utilization percentage, shipment count or customer acceptance data accompanied this October 22 statement.","evidence":"reported","as_of":"2026-10-22","source_ids":["optical-operations-cnr-2026-10-22"]}]
- source_ids: ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"] → ["innolight","innolight-hk-listing-2026-07-29","cioe-optics-2026-09-11","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","optical-operations-cnr-2026-10-22"]
- reviewed_at: 2026-10-01 → 2026-10-23
- Updated: RoboTechnik
4 field changes
- facts: [{"label":"Silicon-photonics backlog","value":"Company reports RMB 2.0 billion of SiPh assembly and testing equipment backlog at June 30, without customer, margin or acceptance details.","evidence":"vendor","as_of":"2026-06-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Silicon-photonics revenue mix","value":"SiPh assembly and test equipment generated RMB 439.1 million in 2025 and 65.8% of group revenue in the first four months of 2026 after the ficonTEC acquisition.","evidence":"vendor","as_of":"2026-04-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Offering plan","value":"At the maximum offer price, the prospectus estimates HKD 4.961 billion of net proceeds, with 20% planned for R&D and 40% for capacity; final proceeds and execution remain pending.","evidence":"primary","as_of":"2026-09-21","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20"]}] → [{"label":"Silicon-photonics backlog","value":"Company reports RMB 2.0 billion of SiPh assembly and testing equipment backlog at June 30, without customer, margin or acceptance details.","evidence":"vendor","as_of":"2026-06-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Silicon-photonics revenue mix","value":"SiPh assembly and test equipment generated RMB 439.1 million in 2025 and 65.8% of group revenue in the first four months of 2026 after the ficonTEC acquisition.","evidence":"vendor","as_of":"2026-04-30","source_ids":["robotechnik-prospectus-2026-09-21"]},{"label":"Offering plan","value":"At the maximum offer price, the prospectus estimates HKD 4.961 billion of net proceeds, with 20% planned for R&D and 40% for capacity; final proceeds and execution remain pending.","evidence":"primary","as_of":"2026-09-21","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20"]},{"label":"Final base offering terms","value":"The September 28 allotment filing sets 11,876,000 H shares at HKD 436 each, with HKD 5,177.9 million gross and HKD 4,961.3 million net proceeds before any over-allotment exercise. The offer-size adjustment option was not exercised; over-allocation was 1,781,400 shares. This resolves base-offering pricing, not realized equipment capacity or spending.","evidence":"primary","as_of":"2026-09-28","source_ids":["robotechnik-allotment-2026-09-28"]}]
- watch_for: ["Final offering proceeds and realized Suzhou capacity additions","Named-customer acceptance, yield, uptime and delivery economics","Provenance of optics, motion control, cameras and test instruments","Transfer and retention of ficonTEC know-how across Germany, Estonia and China"] → ["Over-allotment outcome, use of proceeds and realized Suzhou capacity additions","Named-customer acceptance, yield, uptime and delivery economics","Provenance of optics, motion control, cameras and test instruments","Transfer and retention of ficonTEC know-how across Germany, Estonia and China"]
- source_ids: ["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20"] → ["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","robotechnik-allotment-2026-09-28"]
- reviewed_at: 2026-09-24 → 2026-10-23
- Added: DeepSeek releases DeepInfer with Ascend 950 support
- Added: InnoLight and Eoptolink report normal operations
- Added: RoboTechnik final offering terms backfilled
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-22 → 2026-10-23
- last_automated_run_at: 2026-10-22 → 2026-10-23
- coverage: Screened all 12 priority-1 companies plus oldest-reviewed RSLaser, Guowang, SMEE and Ligent with overlapping English and Chinese searches for October 14-22 and targeted primary-source backfills. Updated only Huawei, Eoptolink and Ligent; other profile dates remain unchanged. Coverage is partial: search results were sparse or stale for several companies, and private deployments, tenders, production metrics and every benchmark artifact were not exhaustively audited. → Screened all 12 priority-1 companies plus the four oldest-reviewed remaining profiles (RSLaser, Guowang, SMEE and RoboTechnik), using overlapping English and Chinese searches for October 15-23 and targeted primary-source backfills. Reviewed supporting documents for Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik; only those five profile dates advanced. Coverage remains partial: several searches returned stale or sparse results, and private deployments, tenders, production metrics and individual benchmark artifacts were not comprehensively audited.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","robotechnik"],"summary":"Screened 16 companies over October 15-23. Added DeepInfer release evidence after reading the initial README at d585e6542c0571176272d703f03c4807ae901284 and confirming the latest README at 9bd9cf243fc53ef9afbf8f04d391d989117d6ab9; recorded Ascend support separately from NVIDIA-only benchmark plots and service exclusions. Read original CNR reporting on October 22 optical-company operating statements and added one cross-company event; these statements do not provide new shipment volumes. Backfilled RoboTechnik final base-offering terms from HKEX pages 2-4, visually checked proceeds, and preserved the September offering plan; subsequent over-allotment outcome and realized spending remain unverified. Reviewed Guowang university-hosted recruiting evidence without claiming qualified optics deliveries. Only Huawei, InnoLight, Eoptolink, Guowang and RoboTechnik profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, checked through GitHub; reviewed the InferenceX public landing page. No matched current Chinese end-to-end result was verified, but individual artifacts were not exhaustively audited and yesterday's independent Ascend kernel evidence remains intact. NAURA supplier-award details and quantitative optical shipment claims remain follow-ups. Sparse/stale lithography search results do not establish absence of progress. Private deployments, unindexed disclosures, tenders, yields, volume, reliability, power and provenance remain incompletely covered. No historical briefing, correction or dated forecast changed.","source_ids":["deepinfer-release-2026-10-20","optical-operations-cnr-2026-10-22","robotechnik-allotment-2026-09-28","guowang-recruitment-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: DeepInfer initial release: supported platforms, service exclusions and benchmark hardware
- Added: Beijing Guowang campus recruitment presentation
- Added: InnoLight and Eoptolink respond on operating conditions
- Added: Announcement of final offer price and allotment results
2026-10-220859f32 ↗
research: add Ascend microbenchmarks and optical evidence backfills
- Updated: data/benchmarks.yaml
4 field changes
- headline: No verified latest-versus-latest independent comparison → No verified matched frontier end-to-end comparison
- assessment: Follow-up through October 21 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 22 did not verify a matched independent end-to-end comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-21 → 2026-10-22
- studies: [{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}] → [{"title":"Ascend 950PR kernel study","evidence":"measured","date":"2026-10-21","finding":"SemiAnalysis reports a best GEMM kernel reaching 99.7% of its compute ceiling; below the compute-bound ridge, median achieved/attainable performance was 41.5% versus H100's 46.8%. Power and clock limits differed. High kernel utilization is not model utilization. Its EP16 communication comparison with GB300 includes different operator work, precluding an unqualified speedup.","source_ids":["semianalysis-ascend-950pr-2026-10-21"]},{"title":"Ascend 910 scientific-computing study","evidence":"measured","date":"2026-07-22","finding":"Pengcheng Laboratory measured Ascend 910A/B/C across scientific kernels and five applications. In one matched quantum-simulation case, a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on A800 with cuQuantum. The study also reports weaker vector and memory-bound behavior, relies on workload-specific optimization, compares against export-limited A800/H800 rather than current frontier chips, and does not establish AI model-training or inference parity.","source_ids":["ascend-science-2026-07-22"]},{"title":"Biren GLM-5.3-Flash functional validation","evidence":"vendor","date":"2026-09-07","finding":"Vendor reports functional and accuracy checks on Bili 166M with SGLang and BIRENSUPA. Compatibility evidence only; no matched foreign-hardware throughput, latency or power comparison.","source_ids":["biren-glm-2026-09-07"]},{"title":"MLPerf Training v6.0 / Inference v6.1","evidence":"standardized","date":"2026","finding":"Standardized submitter-run results. The current Training v6.0 and Inference v6.1 submitter listings do not establish a matched latest Chinese-versus-foreign comparison.","source_ids":["mlperf-training","mlperf-inference-6-1"]},{"title":"InferenceX","evidence":"measured","date":"Continuous","finding":"Reproducible independent inference runs. A June 9 article examined a 16-rank Ascend 950DT DeepSeek V4 profiler trace and verified working day-one inference, but provided no public same-harness numerical comparison with H200 or B200; the promised follow-up was not found. The current dashboard names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems, without a verified matched current Chinese result.","source_ids":["inferencex","inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"title":"Ascend operational field study","evidence":"measured","date":"2026-07","finding":"A small independent preprint documents deployment friction on older Ascend 910 devices. No matched NVIDIA control.","source_ids":["ascend-field"]},{"title":"HiFloat4 pretraining","evidence":"vendor","date":"2026-04","finding":"Demonstrates training progress on dense and MoE models; does not measure frontier time-to-quality against foreign hardware.","source_ids":["hifloat4"]}]
- Updated: Eoptolink
3 field changes
- facts: [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]}] → [{"label":"1.6T shipment progression","value":"Company reports higher second-quarter than first-quarter 2026 shipments; faster second-half growth is an expectation, not a realized result.","evidence":"reported","as_of":"2026-09-11","source_ids":["cioe-optics-2026-09-11"]},{"label":"400G module component inspection","value":"An independent teardown of one Eoptolink 400G DR4+2km module identified a Marvell IN010C50 DSP and multiple Analog Devices power and analog parts; the module's manufacturing date and representativeness were not established.","evidence":"measured","as_of":"2026-09-19","source_ids":["eoptolink-400g-teardown-2026-09-19"]},{"label":"Proposed US procurement restriction","value":"A bipartisan Senate bill introduced on September 25 would prohibit federal procurement of covered Eoptolink optical transceivers for national-security systems after a five-year transition, with limited waivers. It is proposed legislation, not a current or market-wide ban.","evidence":"primary","as_of":"2026-09-25","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"]},{"label":"NPO delivery outlook and capacity","value":"Company expects NPO deliveries from 2027 and larger scale in 2028, conditional on demand. It reports ongoing Chengdu and Thailand capacity expansion; qualified capacity, yields and named customers are undisclosed.","evidence":"vendor","as_of":"2026-09-11","source_ids":["eoptolink-ir-2026-09-11"]}]
- source_ids: ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25"] → ["eop-dsp","eoptolink","cioe-optics-2026-09-11","eoptolink-400g-teardown-2026-09-19","us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","eoptolink-ir-2026-09-11"]
- reviewed_at: 2026-09-29 → 2026-10-22
- Updated: Huawei
3 field changes
- facts: [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]}] → [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]},{"label":"Independent Ascend 950PR kernel measurements","value":"SemiAnalysis tested a 950PR_9579 with 28 Cube cores and 128 GiB HBM. These measurements do not establish end-to-end model parity.","evidence":"measured","as_of":"2026-10-21","source_ids":["semianalysis-ascend-950pr-2026-10-21"]}]
- source_ids: ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19"] → ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19","semianalysis-ascend-950pr-2026-10-21"]
- reviewed_at: 2026-10-21 → 2026-10-22
- Updated: Ligent Technologies
4 field changes
- chinese_name: → 纳真科技
- facts: [{"label":"High-speed product portfolio","value":"Company lists 400G, 800G and 1.6T pluggable transceiver families; the site does not establish shipment mix, customer qualification or production yield.","evidence":"vendor","as_of":"2026-09-22","source_ids":["ligent"]},{"label":"First-half 2026 scale","value":"Reuters reports RMB 5.39 billion revenue and RMB 661 million net profit for the six months ended June 30, up 27.9% and 29.7% year over year.","evidence":"reported","as_of":"2026-06-30","source_ids":["ligent-ipo-2026-09-22"]}] → [{"label":"High-speed product portfolio","value":"Company lists 400G, 800G and 1.6T pluggable transceiver families; the site does not establish shipment mix, customer qualification or production yield.","evidence":"vendor","as_of":"2026-09-22","source_ids":["ligent"]},{"label":"First-half 2026 scale","value":"Reuters reports RMB 5.39 billion revenue and RMB 661 million net profit for the six months ended June 30, up 27.9% and 29.7% year over year.","evidence":"reported","as_of":"2026-06-30","source_ids":["ligent-ipo-2026-09-22"]},{"label":"External light-source commercialization boundary","value":"Ligent reports initial commercialization of 100 mW ELSFP using self-developed CW lasers, while 400 mW ELSFP was scheduled for demonstration. It announced 6.4T/3.2T NPO and 12.8T TFLN XPO products; customer acceptance, shipment volume and component provenance remain unknown.","evidence":"vendor","as_of":"2026-09-07","source_ids":["ligent-cioe-2026-09-07"]}]
- source_ids: ["ligent","ligent-ipo-2026-09-22"] → ["ligent","ligent-ipo-2026-09-22","ligent-cioe-2026-09-07"]
- reviewed_at: 2026-09-22 → 2026-10-22
- Added: SemiAnalysis publishes Ascend 950PR microbenchmarks
- Added: Eoptolink outlines NPO delivery targets and two-country expansion
- Added: Ligent distinguishes commercial light sources from higher-power demonstrations
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-21 → 2026-10-22
- last_automated_run_at: 2026-10-21 → 2026-10-22
- coverage: Screened all 12 priority-1 companies and the four oldest-reviewed remaining profiles using overlapping English and Chinese searches for October 13-21, with targeted filing backfills. Updated Huawei, Skyverse and Enflame only; search-only screening did not advance other review dates. Added reported training plans, customer-validation evidence and an explicit model-name correction. Coverage is partial: the original DeepSeek report was paywalled, NAURA's linked WeChat release could not be read, and private deployments, unindexed disclosures, tenders and individual benchmark artifacts were not exhaustively audited. No matched current Chinese accelerator comparison was verified and no dated forecast or historical briefing changed. → Screened all 12 priority-1 companies plus oldest-reviewed RSLaser, Guowang, SMEE and Ligent with overlapping English and Chinese searches for October 14-22 and targeted primary-source backfills. Updated only Huawei, Eoptolink and Ligent; other profile dates remain unchanged. Coverage is partial: search results were sparse or stale for several companies, and private deployments, tenders, production metrics and every benchmark artifact were not exhaustively audited.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-22","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","smee","ligent"],"summary":"Screened priority-1 companies and the four oldest-reviewed remaining profiles over October 14-22 (seven-day overlap from the preceding run). Added a new independent microbenchmark study, distinguished kernel measurements from vendor-trace analysis and full-model comparisons, and backfilled two optical-company primary documents. Eoptolink's September investor record resolves the earlier NPO roadmap follow-up; Ligent's release distinguishes initial commercialization from demonstrations. Only Huawei, Eoptolink and Ligent profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f, verified through GitHub. InferenceX's public dashboard was reviewed; no matched current Chinese end-to-end result was verified. Individual result artifacts and plotted microbenchmark values were not exhaustively audited. NAURA supplier-award details and newer Eoptolink shipment claims remain follow-ups; neither was promoted to a fact. Search coverage was limited by sparse/stale indexed results, especially lithography suppliers, and does not exclude undisclosed progress. No dated forecast or historical briefing changed. Private deployments, unindexed disclosures, tenders, customer acceptance, yield, volume, utilization, reliability, power and component provenance remain incomplete.","source_ids":["semianalysis-ascend-950pr-2026-10-21","eoptolink-ir-2026-09-11","ligent-cioe-2026-09-07","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: Investor relations activity record 2026-005, September 11
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-21 → 2026-10-22
- Added: CIOE 2026 optical interconnect portfolio and external light sources
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-21 → 2026-10-22
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-21 → 2026-10-22
- Added: The Huawei-DeepSeek Symbiosis: Ascend 950PR microbenchmarks
2026-10-21a16d56f ↗
research: record October 21 adoption evidence and correction
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 20 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads and submitter listings plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 21 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads (unchanged since the previous submitter review) plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-20 → 2026-10-21
- Updated: Enflame
5 field changes
- assessment: TechInsights found TSMC N6 in the examined S60 compute chiplet; this does not identify the provenance of every later product. Enflame displayed an ESL64-0 supernode with an orthogonal, cable-free electrical interconnect and optical scale-out, but supplied no card count, operating deployment, workload, reliability, power or matched performance result. → TechInsights found TSMC N6 in the examined S60 compute chiplet; this does not identify the provenance of every later product. Enflame displayed an ESL64-O supernode with an orthogonal, cable-free electrical interconnect and optical scale-out, but supplied no card count, operating deployment, workload, reliability, power or matched performance result.
- facts: [{"label":"ESL64-0 supernode","value":"Enflame displayed an ESL64-0 supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}] → [{"label":"System partner and corrected model name","value":"Enflame's October 16 release, republished October 18, identifies the model as ESL64-O (letter O) and names ZTE as co-developer. Its claim of supporting clusters above 10,000 cards describes design capability, not a measured or accepted deployment. See the appended correction to the earlier exhibition record.","evidence":"vendor","as_of":"2026-10-16","source_ids":["enflame-ciif-release-2026-10-16"]},{"label":"ESL64-O supernode","value":"Enflame displayed an ESL64-O supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}]
- watch_for: ["Adoption beyond Tencent","Manufacturing route for new generations","Independent training and inference results","Operating ESL64-0 deployments with card count, topology, utilization, reliability and power data"] → ["Adoption beyond Tencent","Manufacturing route for new generations","Independent training and inference results","Operating ESL64-O deployments with card count, topology, utilization, reliability and power data"]
- source_ids: ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12"] → ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12","enflame-ciif-release-2026-10-16"]
- reviewed_at: 2026-10-20 → 2026-10-21
- Updated: Huawei
3 field changes
- facts: [{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]}] → [{"label":"Reported DeepSeek training plans","value":"Investing.com, citing The Information, reports DeepSeek input into future Huawei chip designs, planned Ascend 950DT training as early as end-2026, and priority access to most 960DT supply expected in Q1 2027. These are reported intentions, not verified deliveries or completed training. Contracts, allocation counts, time-to-quality and reliability are unknown.","evidence":"reported","as_of":"2026-10-19","source_ids":["deepseek-huawei-design-investing-2026-10-19"]},{"label":"Reported first-half shipment scale","value":"IDC data reported by Reuters put Huawei at 572,000 AI accelerators shipped in China in the first half of 2026, up from more than 300,000 a year earlier and equal to 28% market share. Product mix, buyers, accepted deliveries, utilization and workload results were not disclosed.","evidence":"reported","as_of":"2026-06-30","source_ids":["china-ai-accelerator-share-reuters-2026-10-13"]},{"label":"Cross-border patent licensing","value":"Huawei and Qualcomm announced a multi-year cross-license spanning 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents. The transaction remains subject to regulatory approvals; terms, covered patents and their use in specific products were not disclosed, so it does not establish a hardware-supply dependency.","evidence":"primary","as_of":"2026-10-05","source_ids":["huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05"]},{"label":"Independent Ascend 950DT trace analysis","value":"InferenceX examined a 16-rank DeepSeek V4 decode profile on Ascend 950DT and found working, optimized day-one inference with overlapped compute and communications. The article did not publish a same-harness numerical Ascend-versus-H200/B200 result and said that comparison was future work; its profiler images are attributed to SemiAnalysis and Huawei.","evidence":"measured","as_of":"2026-06-09","source_ids":["inferencex-ascend-950dt-deepseek-v4-2026-06-09"]},{"label":"Ascend 950 cloud-service evidence","value":"Huawei announced its Ascend 950 cloud cluster service and scheduled Chinese commercial availability for September 30, claiming 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior compute generation. Configuration, model, quality, customer use and independent availability or performance confirmation were not disclosed.","evidence":"vendor","as_of":"2026-09-18","source_ids":["huawei-cloud-ascend-950-service-2026-09-18"]},{"label":"Open-source Ascend 950 programming path","value":"DeepSeek contributors published TileLang's native Ascend 950 backend with code generation, scheduling, synchronization and kernel examples; Reuters reports related compute and communication libraries for a 128-chip Ascend 950 supernode. This is inspectable software evidence, not a matched workload result or proof of CUDA equivalence.","evidence":"primary","as_of":"2026-09-30","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30"]},{"label":"Hi-ONE optical-interconnect readiness","value":"Huawei says its 7.2T Hi-ONE near-packaged-optics engine has mass-production capability; shipped volume, component provenance, yield, customer acceptance and independent reliability data were not disclosed.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-hi-one-2026-09-17"]},{"label":"Atlas system development status","value":"Huawei says a 256,000-card Atlas 950 cluster is being deployed and a near-packaged-optics Atlas 960 system is under test; neither is a verified completed deployment or workload result.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Peerium interconnect design","value":"Huawei describes UnifiedBus as a common protocol across compute, memory, storage and networking, with unified memory addressing and nested parallelism; independent scaling data are absent.","evidence":"vendor","as_of":"2026-09-17","source_ids":["huawei-peerium-2026-09-17"]},{"label":"Scientific-computing measurements","value":"A Pengcheng Laboratory preprint measured Ascend 910A/B/C on five scientific workloads; a tuned 910C quantum-simulation run took 11.4 seconds versus 14.3 seconds on A800, while vector- and memory-bound kernels remained constrained. This is not a model-training or inference result.","evidence":"measured","as_of":"2026-07-22","source_ids":["ascend-science-2026-07-22"]},{"label":"Domestic supply constraint","value":"Huawei executives say it cannot produce enough AI computing equipment to meet Chinese demand, Ascend competes with smartphone chips for China's very limited advanced-semiconductor capacity, and China still relies on DUV while working toward EUV. No capacity, order, fulfillment, yield or equipment-provenance figures were disclosed.","evidence":"reported","as_of":"2026-10-01","source_ids":["huawei-connect-2026-09-17","huawei-advanced-capacity-reuters-2026-10-01"]},{"label":"Ascend 950 commercialization claim","value":"Huawei materials say Ascend 950 systems entered commercial use; configurations, volume, customer acceptance and workload results were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Scale-up architecture roadmap","value":"Huawei says Peerium is intended to link up to one million processors and an Ascend 960 supernode is planned for up to 4,096 processors; these are architecture targets, not measured results.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"Reported supernode shipments","value":"Huawei says it shipped more than 1,000 supernodes to more than 370 customers; customer, configuration, utilization and workload details were not disclosed.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"2027 accelerator roadmap","value":"Huawei expects the 960DT in Q1 2027 and Ascend 960PR in Q3 2027; these are launch targets, not delivered-product evidence.","evidence":"reported","as_of":"2026-09-17","source_ids":["huawei-connect-2026-09-17"]},{"label":"China Unicom deployment","value":"Huawei says China Unicom deployed Ascend compute at scale and a 384-NPU supernode, and is adapting Ascend 950 for Unicom Cloud; independent operating metrics were not disclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["huawei-china-unicom-2026-09-10"]},{"label":"PyTorch accelerator integration","value":"Huawei is a PyTorch Foundation Platinum member and co-chairs its Accelerator Integration Working Group; governance participation does not prove workload parity.","evidence":"primary","as_of":"2026-09-09","source_ids":["pytorch-china-2026-09-09"]},{"label":"Atlas 350 / 950PR","value":"112 GB HBM; 1.4 TB/s; 425 TFLOPS BF16; up to 600 W","evidence":"vendor","as_of":"2026-09-11","source_ids":["atlas350"]}]
- source_ids: ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13"] → ["ascend","hbm-reuters","pytorch-china-2026-09-09","huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","ascend-science-2026-07-22","huawei-peerium-2026-09-17","huawei-hi-one-2026-09-17","tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","huawei-cloud-ascend-950-service-2026-09-18","huawei-advanced-capacity-reuters-2026-10-01","inferencex-ascend-950dt-deepseek-v4-2026-06-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","china-ai-accelerator-share-reuters-2026-10-13","deepseek-huawei-design-investing-2026-10-19"]
- reviewed_at: 2026-10-13 → 2026-10-21
- Updated: Skyverse
4 field changes
- assessment: A product catalog does not establish parity with the most sensitive frontier instruments. → Company filings add customer-validation and sales evidence beyond the catalog. They do not establish frontier sensitivity, throughput, yield contribution or a fully domestic bill of materials.
- facts: [] → [{"label":"Customer validation and sales","value":"Company says 11 of its 13 equipment families passed leading domestic customer validation and generated sales. Mature-process bright-field and X-ray TSV copper-void tools passed validation; HBM wafer-flatness tools remained under evaluation, and electron-beam critical-dimension tools had shipped for validation. Customers, nodes, unit volumes and operating metrics were undisclosed.","evidence":"vendor","as_of":"2026-09-10","source_ids":["skyverse-ir-2026-09-11"]}]
- source_ids: ["skyverse"] → ["skyverse","skyverse-ir-2026-09-11"]
- reviewed_at: 2026-09-17 → 2026-10-21
- Added: Reported DeepSeek–Huawei plans sharpen the training milestone
- Added: Correction: Enflame supernode is ESL64-O, with letter O
- Added: Skyverse filing separates validated tools from ongoing evaluations
- Updated: data/research-state.yaml
5 field changes
- last_research_at: 2026-10-20 → 2026-10-21
- last_automated_run_at: 2026-10-20 → 2026-10-21
- last_run_status: updated → partial
- coverage: Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, plus targeted backfill searches. Added CXMT's reported 4F2 server-memory roadmap, Enflame's displayed ESL64-0 supernode and AMEC's reported installed reaction-chamber scale and two memory-equipment launches. These disclosures establish announced architecture, integration and commercial-scale evidence but not completed 4F2 products, operating supernode deployments, new-tool customer qualification, workload performance, yield or provenance. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 and Inference v6.1 repository heads were unchanged, and InferenceX still supplied no matched current Chinese accelerator result. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance. → Screened all 12 priority-1 companies and the four oldest-reviewed remaining profiles using overlapping English and Chinese searches for October 13-21, with targeted filing backfills. Updated Huawei, Skyverse and Enflame only; search-only screening did not advance other review dates. Added reported training plans, customer-validation evidence and an explicit model-name correction. Coverage is partial: the original DeepSeek report was paywalled, NAURA's linked WeChat release could not be read, and private deployments, unindexed disclosures, tenders and individual benchmark artifacts were not exhaustively audited. No matched current Chinese accelerator comparison was verified and no dated forecast or historical briefing changed.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-21","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened priority-1 Aishengna, AMEC, Cambricon, CXMT, Enflame, Eoptolink, Huawei, InnoLight, NAURA, SMIC, YMTC and Yuliangsheng, plus oldest-reviewed RSLaser, Guowang, Skyverse and SMEE. The overlap began October 13 (seven days before the preceding run). Backfilled Skyverse's original September investor document from the SSE investor platform, distinguishing validated and sold tool families from ongoing HBM and electron-beam evaluations. Recorded Huawei adoption intentions from accessible relay reporting; The Information original remained paywalled, so corroboration is incomplete. Corrected Enflame's model-name spelling in the current profile and appended a correction event while retaining the original news. Only Huawei, Skyverse and Enflame profile dates advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; Inference closed-directory web retrieval failed, but the unchanged repository head was verified through GitHub. InferenceX's current public coverage did not establish a matched current Chinese accelerator result. Individual artifacts were not exhaustively audited. NAURA's supplier-award detail and Eoptolink's newer shipment/roadmap claims need primary-document follow-up. No dated forecast or historical briefing changed. Search-only profiles retain their previous dates; coverage remains partial for private deployments, unindexed Chinese disclosures, tenders, production yield, volume, acceptance, utilization, reliability, power and component provenance.","source_ids":["skyverse-ir-2026-09-11","deepseek-huawei-design-investing-2026-10-19","deepseek-huawei-design-information-2026-10-19","enflame-ciif-release-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}]
- Added: DeepSeek–Huawei chip-design reporting
- Added: DeepSeek teams up with Huawei to challenge Nvidia
- Added: Enflame CIIF release: ESL64-O and ZTE collaboration
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-20 → 2026-10-21
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-20 → 2026-10-21
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-20 → 2026-10-21
- Added: Skyverse investor activity record, August 31–September 10, 2026
2026-10-20d104531 ↗
fix: restore complete research-state upload
- Updated: data/research-state.yaml
9 field changes
- : Y��x-���jם��i��+��j[h��ܢ��ߎ9�:-jZ.����)W66�V��fW'6��� �&6VƖ�U�FFS�##b�����7E�&W6V&6��C�##b��# ��7E�WF��FVE�'V��C�##b��# ��7E�'V��7FGW3�WFFV@�6�fW&vS��Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2�#��W2F&vWFVB&6�f���6V&6�W2�FFVB5��Bw2&W�'FVBDc"6W'fW"��V��'�&�F��V�f��Rw2F�7��VBU4�cB�7WW&��FR�B�T2w0�&W�'FVB��7F��VB&V7F����6��&W"66�R�BGv��V��'��WV��V�B�V�6�W2�F�W6RF�66��7W&W0�W7F&Ɨ6����V�6VB&6��FV7GW&R���FVw&F����B6���W&6���66�RWf�FV�6R'WB��B6���WFV@�Dc"&�GV7G2��W&F��r7WW&��FRFW����V�G2��Wr�F���7W7F��W"VƖf�6F����v�&���@�W&f�&��6R���V�B�"&�fV��6R���ǒ5��B�V�f��R�B�T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb��B��fW&V�6Rcb�&W�6�F�'��VG2vW&RV�6��vVB��B��fW&V�6U�7F����7WƖVB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�B���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2���F�f�GV�&V�6��&�'F�f7G2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B��WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�F��W���S�WW&�R����F��'V�3���FFS�##b��� �7FGW3�'F���6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��6Т�&�&V��6��� �7V��'���ТF�W6RR6���W2vW&R67&VV�VB���B��gV�ǒ&WfƖFFVB�&VB�&�v���4��R&W�'F��p��B�F�6�7WƖW'2rV&Ɩ26�FW3�FFVB&W�'FVB6���V�B&�w&W76���f�"����Ɩv�@��BV�F�Ɩ��&VB&�&V�w2FFVB&��'����V�6V�V�G3�FFVB6�gGv&R�fƖFF����@�&W6V&6���&�&F�'�WfV�G2�&W6W'f��rF�RF�7F��7F���&WGvVV�V&Ɩ6F����BWfV�BFFW2�&Wf�WvVBF�R��FW�VB�&�v���6WFV�&W"&WWFW'2V�f��R&W�'B�BFFVB���VBW6P��b��&�6VVG2v�F��WBGWƖ6F��r�G2W��7F��r��WfV�B���ǒF�W6Rf�W"6���&Wf�Wp�FFW2Gf�6VB�&��&�G��6���6V&6�W2�6�&Wf�6�FVBF�R6WFV�&W"�$�&�6��p�&W�'B�B�VǒEUb&W�'F��s���FF�F����6���W&f�&��6R�"VƖf�VB�F���6���v0�66WFVB��T2w2�ff�6��6�FR�BF�RgV��f���6��F��W2EUb'F�6�R6�V�B��B&R&WG&�WfVB�f����rW�T2w26���F������V�6V�V�B�F�REUbFW7F��r6�&�����w��&W�'FVB�F�6����GV�P�W��'B&W7G&�7F���2��B�V��'��WV��V�Bf�Ɩ�w3�F���B��fW"VƖf�6F���g&���VFƖ�W2�F�v���FVB�Bf�7V�ǒ6�V6�VB&�&V�w2&��'���bזV"��f�w&��3�FFVB�G2&WfV�VR����72�"dB�BV���VB�7W7F��W"FVƗfW'�6���2v�F�fV�F�"GG&�'WF������B���FWV�FV�BVF�B�4��B4���6V&6�W2F�B��BW7F&Ɨ6�fW&�f�VB�WrVƖg���r֖�W7F��S�F�V�"&�f��P�FFW2&V���V�6��vVB�6�V6�VB��W&bG&���rcb�&W�6�F�'�Ɨ7F��w2���fW&V�6Rcb� �6��6VB�7V&֗GFW"Ɨ7F��w2f�v�D�V"��B��fW&V�6U�w2V&Ɩ26�fW&vR����F6�VB7W'&V�@�6����fW'7W2�g&��F�W"&W7V�Bv2fW&�f�VC���F�f�GV�&W7V�B'F�f7G2vW&R��BW��W7F�fVǐ�VF�FVB���7F�&�6��Ww2�W&��B'&�Vf��w2�f�&V67G2�B66V�W&F�"7V6�f�6F���2&R&WF��VB�&W6V&6�66�R�2'F����FWV�FV�Fǒ�bv�D�V"7��2�"6��VFf�&RV&Ɩ6F���7FGW2�6�W&6U��G3���6��R��F�72�##b�����&�&V��##bփ��&�&V��v���##b���p��&�&V��6��v����"�##b��� ��V�f��R֗���&��&WWFW'0��GWb�&WWFW'0����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���0�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��6Т�6��� ��V��&V�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�RF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W2�FFVB�F�&6�f�V�FF���&��'�Wf�FV�6RF�@�6�'&�6������VB2�F��V��V�&W"v�F�v�fW&��r&�&B�BFV6��6�Gf�6�'�6�V�6���&W&W6V�FF����B&W�'FVBW7G&V�v�&�7&�726����R�'V�F��R�BF�7G&�'WFVB&V2�FFVB�VvV�w2W��7F��r�F��V��V�&W'6���B6��6��"&��R��F�Rf�V�FF���w266V�W&F� ���FVw&F���v�&���rw&�W�F�W6R&RV6�7�7FV��Bv�fW&��6R6�v��2���BW&f�&��6R&W7V�G2�6��6VBF�R&��"�T2f����r�WW6��rFWF��VB66�V�6R�BFV6�����w�F�ǒ&W�'F��r��6���54T2�V�6�W27&�72WF6��T5dB��B��WF��f���FW�6�F����B6�Ɩ6���6&&�FRW�F�����WB7V6�f�6F���2F�7F��7Bg&��7W7F��W"VƖf�6F����Bf��V�R�&�GV7F���Wf�FV�6R�&Wf�WvVB7W'&V�B�ff�6���'Ff�Ɩ��B�Ww2vW2f�"4�&W6V&6�6��v���4����@�V��&V����WrVƖg���r&�GV7F����7W7F��W"�"6���WFR�f��r֖�W7F��Rv2fW&�f�VB��6�F�V�"f7G2vW&RV�6��vVBv���RF�V�"7GV�&Wf�WrFFW2Gf�6VB�&V6�V6�VBF�P�f���6��F��W2EUb7F�'���$�&�6��r6�fW&vR�F�R6WFV�&W"�F�72&W�'B�BV�f��Rw0���f����r�WGW&��rF�R�fW&�v��F�s���FF�F����6���v266WFVB�&Wf�WvVBF�Rd42w0�f���6����V�B�'V�R6�fW&vR'WBF�B��BGF6��V�F�G��7V6�f�2&W7G&�7F���F�����Ɩv�@��"V�F�Ɩ�&V6W6RF�R'V�RF�B��B��RF�V����W&bG&���r�B��fW&V�6Rcb�&W�6�F�'���VG2vW&RV�6��vVB��BF�R6��6VB��fW&V�6R7V&֗GFW"Ɨ7B7F���6��vVB���F6�VB7W'&V�@�6���W6R66V�W&F�"���fW&V�6U�6�fW&vR7F���F�B��B7Wǒ�WV�f�V�B6���W6R&W7V�B�����FWV�FV�B�FW7B�fW'7W2��FW7B&�G�6���v2FFVB���7F�&�6��Ww2�&W�'G2��f�&V67G2�Wf�FV�6R�&V�2�B&��"'V�&V6�&G2vW&R&W6W'fVB�6�W&6U��G3����F�&6��6�'&�6���##b���p���F�&6��6����##b�������V2�76V2�##b���P��6Т�6��� ��V��&V��GWb�g@���&��&WWFW'0��6��R��F�72�##b�����V�f��R֗����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���@�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��wV�v�p���60���vG6��p�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F�6WFV�&W"~( 3B�fW&���rV�vƗ6��B6���W6R6V&6�W2�FFVB�42&��'�Wf�FV�6P�f�"�G2&W�'FVB������w6�RDD2FW����V�B�r�Ɩ��B�D42'F�6�F����U����GV�P�7W�'B�B�V��F�6�֖�FW&6���V7BF�&V7F����F�RF�66��7W&RF�B��B&�f�FR�F6�V@���f��&�Bv�&���B&V�6��&��6�W7FW"F����w��7v�F6��6�Ɩ6��&�fV��6R�"��FWV�FV�@�7W7F��W"FF�FFVB&W�'F��r6�W&6VBF��vG6��rw26������V�6V�V�BF�B�G0��7FW"�SU��V���WfV�4�7�7FV�v26��VBF��Gf�6VB�6�v��r&�GV7F���Ɩ�S���WB6���V�BF�7F��7Bg&��7W7F��W"66WF�6R���V�B�WF��R�F�&�Vv�WB�Bf��V�P�Wf�FV�6R�&V�V�VBwV�v�rw2FV�B&V6�&B�BF�R7W'&V�B�42�B�vG6��r6���vW3��wV�v�rv��VB���Wr&�GV7F���Wf�FV�6R�6V&6����ǒ6�fW&vR�bF�R�F�W"&��&�G��&�f��W2F�B��BGf�6RF�V�"&Wf�WrFFW2�&V6�V6�VB��W&bG&���r�B��fW&V�6Rcb� �&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR�F�P���W&b&W�6�F�&�W2�B���WvW"&W7V�BW6��B��7W'&V�B6���W6R66V�W&F�"7V&֗GFW �v2f�V�C���fW&V�6U�Ɨ7FVB7W'&V�Bt#3�ԓ3SU��BERcr7�7FV�2'WB��WV�f�V�@�7W'&V�B6���W6R�&Gv&R&W7V�B����F6�VB�FW7B�fW'7W2��FW7B&�G�6���v2FFVB���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�Wf�FV�6R�&V�2�B&��"'V�&V6�&G2vW&P�&W6W'fVB�6�W&6U��G3����62��G��##b���@���vG6��r��V��6��##b������wV�v�p���&��&WWFW'0��6��R��F�72�##b�����V�f��R֗����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���@�7FGW3�WFFV@�6���W3����6�V�v����V0��&�&V��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV����vG6��p����WfF ������Ɩv�@���WF����W&��'6�6W ��6֖0����F0���VƖ�w6�V�p�������7V��'���Т&�F�RvVV�ǒ�V�W"f�"F�R&Wf��W2gV�ǒ6��6VB���F��7V�F�W&��B�B&WF��VBF�P�6WFV�&W"W&��B�F��FFRVF�F���V�6��vVB�VF�FVBF�RFV�6V�V7FVB���W&��BWfV�G2�@�&��fW&���rV�vƗ6��B6���W6Rv6V&6�W27&�7266V�W&F�'2�6�gGv&R��V��'����F�6���FW&6���V7G2�ƗF��w&���B6�v��rWV��V�B�FFVB&�&V�w26WFV�&W"rfV�F� �66�V�B�b�VwW7B#�V6�7�7FV��VWGW26���6�6�W&6R�BWfV�C��B6�2F�&6�5U�W6W2�F�&6�&�fFUW6S�B7W�'G26������2�6���V7F�fW2�Be4E"�w&�6GW&R�@�&�f�Ɩ�r���V&Ɩ26�FR&W�6�F�'�v2�FV�F�f�VB�6�F�R&V6�&BF�W2��B��fW"�W&F� �6�fW&vR����F��&�ƗG��W7G&V�66WF�6R�W&f�&��6R�"5TD&�G��F�R6���WFP�'&�Vf��r&W6W'fW2����7F�&�6�&W�'G2�B7FFW2F�B��FFVBf�&V67B��fVB��BW6W0�F�R7W'&V�B6WFV�&W"B&V�6��&�&Wf�Ws����F6�VB6��FV��&'�6���W6R&W7V�Bv��7@�t#3�ԓ3SU��"ERcrv2fW&�f�VB����W&bcb��"��fW&V�6U�6�fW&vR�&V6W6R7�7FV�F�0�6�fW&vR&Vv�6WFV�&W"�F�R6WFV�&W"r�2&V6��7G'V7F���&V���2��6���WFRf�"&�fFP�FW����V�G2�V��FW�VBF�66��7W&W2�FV�FW'2�B��F�f�GV�&V�6��&�'F�f7G2�6V&6����ǐ�6�fW&vRF�B��BGf�6R�F�W"6���&Wf�WrFFW2�6�W&6U��G3���&�&V��F�&6�7W�##b���p��&�&V��##bփ��&�&V��v���##b���p��&�&V��6��v����"�##b��� ���F�&6��6�'&�6���##b���p���F�&6��6����##b�������&��&WWFW'0��6��R��F�72�##b�����V�f��R֗���vG6��r��V��6��##b������GWb�g@����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���P�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p����WfF ���6W@�����w6V�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"��R����WrVƖg���p����v��F�r66V�W&F�"��V��'���F�6�֖�FW&6���V7B�"6V֖6��GV7F�"�WV��V�BWfV�Bv0�fW&�f�VB�f�Ɩ�r�v&Wf�Wrf�V�BF�&VR�FW&��VwW7BF�66��7W&W2F�BvW&R֗76��rg&�ТF�R6���6�&V6�&B��T2w26���&W�'B6�2F�&VR�5WF6�f֖ƖW2vW&R��f��V�P�&�GV7F�����Ɩ�W2B��&RF��S7W7F��W'2�v�F�7F&�R&�GV7F���W6R�"&WVB�&FW'2f� �FF�F����Gf�6VBWF6�7�7FV�2�B&WVB&�GV7F����&FW'2gFW"��6�FRfƖFF����`�GV�w7FV�FW�6�F���7�7FV�2B�W��V��'��B��v�27W7F��W'2��U$&W�'G2��&RF��� �7V�V�F�fRFVƗfW&�W2V6�f�"WF6��dB�BfW'F�6�gW&�6W2�$�"#�c&��Ɩ���bf�'7Bֆ�`�&WfV�VR��B&����WB�b��'&�B�&��F��r�B��v��V�BE5bV�V7G&��F��rF���2��4UB&W�'G0��Dd��f��V�R&�GV7F����7W7F��W"fƖFF����b6�Ɩ6�����F��72V�v��R6��W2��BC�BP��V"��fW"זV"w&�wF���6��WF��r�V�V7G&��72&WfV�VR�V6�6���&V���2fV�F�"Wf�FV�6S��V���VB7W7F��W'2�V�7V6�f�VB&�6W72��FW2��BV�F�66��6VB��V�B�66�G��WF��R��F�&�Vv�WB�"&�GV7F���f��V�RƖ֗BF�R6��6�W6���2�&Wf�WvVB��WfF"�B���w6V֒w0�7W'&V�B�ff�6��vW2v�F��WB66WF��r�Wr&�GV7F����"&V�6��&�6����&V6�V6�V@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vS�����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU��"ERcr�6�727�7FV�0�v2f�V�B�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R&Wf�WrFFW2���7F�&�6�&W�'G2��6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVC�F�RVwW7BWfV�G2&RW�Ɩ6�Fǐ��FR�FFVB&F�W"F��FFVB26WFV�&W"WfV�G2�6�W&6U��G3����V2�##bփ�&W�'@���W&�##bփ�&W�'@���6WB�##bփ�&W�'@����WfF �����w6V����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���`�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p���V��V熖���WF������&R�F�&VG0�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"��b�FFVB���&RF�&VG2p�6WFV�&W"�&��'����V�6V�V�BF�B�B6��VB��2��6&BgV���gV�7F���uR6�W7FW"�F�R&V6�&BG&VG2F��22��&F�W"F��FVƗfW'�Wf�FV�6R&V6W6R��66�VGV�R�uR��FV���6���WFVB6&B6�V�B�F����w����FW&6���V7B��vW"�&V6�fW'��"v�&���B&W7V�Bv2F�66��6VB�&6�f���VB���&RF�&VG2rVwW7B��bזV"&W�'C�6����&W�'FVB&WfV�VRv2$�"�s3`�&��Ɩ���GG&�'WF&�R�WB��72v2$�"�Sb֖�Ɩ����B$�"cc֖�Ɩ���TR6�W26��G&7@�v2gV�ǒFVƗfW&VB�B&V6�v旦VB��G2FV�6R�B��R�eR�66Ɩ�r�Vff�6�V�7��BVffV7F�fRТG&���r�F��Rf�wW&W2&V���fV�F�"�V7W&V�V�G2v�F��WB�F6�VBv�&���B��vW"�"7W'&V�@�f�&V�v�ֆ&Gv&R6��&�6���FFVB�V��V熖�w2F�v���F&�Rf�t�2֖�5�R�$"FW����V�B��vP��B�WF�w2E$�Fv�&Wf�WrF��FFF����W2&W�'FVB�F�&6��W&F�"6�fW&vR0�6��F�&�ƗG�Wf�FV�6R���BW&f�&��6R�"5TD&�G��&V6�V6�VB��W&bG&���r�B��fW&V�6P�cb�&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w27W'&V�B6�fW&vR����F6�V@�6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R&Wf�WrFFW2���7F�&�6�&W�'G2�6�'&V7F���2��f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVC�F�R��bזV"f�Ɩ�r�2W�Ɩ6�Fǒ�FR�FFVB�6�W&6U��G3������&R�F�&VG2�##bփ�&W�'@�����&R�F�&VG2֦B�6��VB�##b�������V��V熖��֖�7��##b��� ���WF��G&��##b���@���F�&VG0���V��V熖����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���p�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�6��fW'6P�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"�r�FFVB&WWFW'2p�6WFV�&W"r66�V�B�b�VvV�w2�Wr�V�6�F&vWG2f�"F�R�cEB��##r�B66V�@��c"��2##r��W2�VvV�w26���F�B�B�B6��VB��&RF���7WW&��FW2F���&P�F��3s7W7F��W'2�F�R&V6�&B&W6W'fW2F�R��FW"FFVB&�F��BG&VG2F�R�WrFFW20�W�V7FVB֖�W7F��W2���BFVƗfW'�Wf�FV�6R��VvV�F�B��B�FV�F�g�7W7F��W'2�7FFRF�P��V�&W"�b66V�W&F�'2W"6��VB7�7FV��"F�66��6R66WF�6R�WF�Ɨ�F����v�&���B���vW"�&VƖ&�ƗG��"��FW&6���V7B�W&f�&��6R&W7V�G2�FFVB�VvV�w26WFV�&W"66�V�@��b6���V�6��FW�����r66V�B6��WFR�B3�B��R7WW&��FR�BFF��r66V�B�Sf� �V�6��6��VC�F��2�2��VB�FW����V�BWf�FV�6R'WB&V���2fV�F�"66�V�Bv�F��WB��FWV�FV�@��W&F��r�WG&�72�&Wf�WvVB6��fW'6Rw27W'&V�B�ff�6���'Ff�Ɩ��B�Ww2vW2v�F��W@�66WF��r&�F�G�R�"&�GV7F���6���2g&����vR���ǒ&W�'F��r�%4�6W"�B4���6V&6�W0�&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�V@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'���B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�@�t#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6P�&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3����VvV��6���V7B�##b���p���VvV��6����V�6���##b��� ��6��fW'6P����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b�����7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�6�VP�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"���FFVB&WWFW'2p�6WFV�&W"�&W�'BF�B5��B��2��B"dB&�GV7F���Ɩ�RB�G2�Wr&V����r��B�@��2F�67W76VB&�7V7F�fR��7F�&vR7W7F��W'2�F�RWfV�B�2&V6�&FVB2&W6V&6��BG&��Т&�GV7F��������B6���W&6���WGWC�&WWFW'26�V�B��BW7F&Ɨ6���W&F��rFFR�"�6��֗F�V�BF�f��V�R��Vf7GW&��r��B5��BF�B��B6���V�B�&R�&Wf�WvVB&WWFW'2r�FW&��ǐ�W��FVB6WFV�&W"r�VvV�&W�'B�BFFVB6�����WfV�Bf�"F�R6���w27FFV�V�G0�V��Vv��6��WF��rWV��V�BF�6F�6g�F��W7F�2FV��B�66V�B�S7�7FV�2�fRV�FW&V@�6���W&6��W6R��66V�B�c7WW&��FR�2���VBf�"WF�B��b&�6W76�'2�BVW&�V��0���FV�FVBWfV�GV�ǒF�6���V7B2��2��R֖�Ɩ���F�W6R&R&W�'FVB6���6���3���VvV�F�B��BF�66��6R66�G���&FW'2�7W7F��W"66WF�6R�WF�Ɨ�F����v�&���B��vW"��&VƖ&�ƗG��"��FW&6���V7B�W&f�&��6RFF�&Wf�WvVB4�TRw27W'&V�B�ff�6��&�GV7BvS�����WrVƖf�VB&�GV7F���֖�W7F��Rv266WFVB�%4�6W"�B4���6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VB��W&`�G&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'��@���fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�@�t#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ6�fW&vRF�B��BGf�6R�F�W �6���&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���7��B���B�##b�������VvV��6���V7B�##b���p��6�VP����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b�����7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�F�V@�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W""���FFVB���FWV�FV�@�FV&F�v��b��RV�F�Ɩ�CrE#B�&����GV�RF�BF�&V7Fǒ�FV�F�f�VB�'fV����3SE5 ��B6WfW&����rFWf�6W2�vW"�B���r'G2�F�R&V6�&BW�6�VFW2F�RWF��"w2��fW'&V@��6W"�FV�F�f�6F����BƖ֗G2F�Rf��F��rF�F��2�����W2�6��v�R6��Rv�F���F�66��6V@���Vf7GW&��rFFS��BF�W2��BW7F&Ɨ6�F�R&����b�FW&��2f�"7W'&V�B�r�"�e@�&�GV7F����F�R��FW7B�&�f��R&Wf�Wr&V�V�VBBԆVBw27W'&V�B�ff�6��vR�BF�RW��7F��p���&WWFW'2&W�'B�F�V�&6�f���VBF�R�FW&��ǒV�FW"�&V6�&FVB6���6���2�b��&RF��Sc���V�wRV�G26��VB�FW����V�G2B��&RF��CW�FW&��7W7F��W'27&�72# ���GW7G&�W2�F�R#��66V�W&F�"榗R�#�&6��BW�V7FVBc��B��&�F�FFW2�&�GV7B֗��66WF�6R�WF�Ɨ�F����v�&���BW&f�&��6R��vW"�f'&�6F����B�V��'��&�fV��6R&V���V�F�66��6VB�%4�6W"�B4���6V&6�W2v��&�GV6VB��&WG&�Wf&�R�Wp�&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vVB��W&bG&���p��B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'��B��fW&V�6U�w0�7W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU���"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W �6���&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���V�F�Ɩ��Cr�FV&F�v��##b������Ӄ� ��F�V@����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"2�#�FFVB&WWFW'2p�6WFV�&W"#66�V�B�b5��Bw27FFV�V�BF�B�G2f�gF��vV�W&F���E$��Ff�&�V�FW&V@��72&�GV7F���W6��rVG'W�RGFW&��r�B��R��7F�fR�&V��b��F6���BF�@�Gv�#Bv"�DE#U�&�GV7G2�FR��F�R�Ff�&�&R���72&�GV7F����F�R&V6�&BG&VG0�F�W6R26���6���2�B&W6W'fW2&WWFW'2rF�7F��7F���&WGvVV�w&�72F�W2W"vfW"�@�F�R&��'F���76��rf���FW7G3�F�R6���VBB��V7B�SR��&�fV�V�B�fW"F�Rf�W'F�ТvV�W&F����Ff�&�F�W2��BF�66��6R��V�B�&VƖ&�ƗG��7W7F��W"VƖf�6F����"�WGW@�f��V�R�5��B6�B6���W6RWV��V�B��W'26��G&�'WFVBF�7&�F�6�&�6W727FW2'WBF�B��@���RF�V��"W7F&Ɨ6�gV�ǒF��W7F�2&�6W72f��r�F�RWfV�B�26����F�G��E$�֖�W7F��R����BWf�FV�6R�b�$�f��V�R�"W&f�&��6R�%4�6W"�4����B����6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vV@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR��v��6���r��W2'V&�����w6�FRt#3�ԓ3SU��BERcr����F6�VB6��FV��&'�6���W6P�66V�W&F�"&W7V�Bv2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W"6��琢&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���7��B�sR�G&��##b���# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"B�#��FR�FFVB�Vǒ# �V�v6�V�r�&�&F�'�&W&��BF�B�V7W&W266V�B��"�2��66�V�F�f�2�W&�V�2�Bf�fP�Ɩ6F���7GVF�W2�F�R&V6�&B6GW&W2ff�&&�R�F6�VBV�GV��6��V�F���66S�GV�V@��2���V�V�FF���6���WFVB3�V&�B�3���W"6�&7V�B���B6V6��G2fW'7W2B�26V6��G0�����W6��r7UV�GV���B�6�&W6W'fW2F�RW"w2�W72ff�&&�RWf�FV�6S�fV7F�"��@��V��'��&�V�B�W&�V�2&V���VB6��7G&��VB�6WfW&�6��&�6��2W6VBF�ffW&V�B�v�&�F��2� ����V�V�FF���2��Bv�&���B�7V6�f�2�F�֗�F���v26V�G&��F�R7GVG�W6W2W��'B�Ɩ֗FV@���6��G&��2&F�W"F��7W'&V�Bg&��F�W"�&Gv&R�BF�W2��BFW7B���FV�G&���r� ���fW&V�6R�6��BF�W2��B6��vRF�R�FW7B�fW'7W2��FW7B76W76�V�B�%4�6W"�4����B�����6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vVB��W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w0�7W'&V�BV&Ɩ26�fW&vS����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU���ERcr�"F�R�WvǒƗ7FVB��V��7�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��@�Gf�6R�F�W"6���&Wf�WrFFW2���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6P��&V�2vW&R&W6W'fVB�6�W&6U��G3���66V�B�66�V�6R�##b�r�# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#�7FGW3�WFFV@�6���W3���7��@��V�F�Ɩ氢��VvV����WF�����F0�7V��'���Т&�F�RvVV�ǒ�V�W"f�"F�R&Wf��W2gV�ǒ6��6VB���F��7V�F�W&��B�B7&VFVBF�R6���WFP�6WFV�&W"B�#VF�F���v�F��WB&W�6��rF�RV&ƖW"6WFV�&W"rW&��B�F��FFR�"6���WFP�&W�'G2�VF�FVB��6��6V�V7FVB���W&��BWfV�G2�BF�V�"�&�v���6�W&6W2�F�V�&�&�V�FV@�V�vƗ6��B6���W6Rv6V&6�W27&�7266V�W&F�'2��V��'���F�6���FW&6���V7G2�6�gGv&R�@���Vf7GW&��rWV��V�B�F�R7��F�W6�2G&VG2�VvV�w266V�W&FVB##r&�F��&W�'FVB7�7FVТ66�R�BVW&�V�F&vWG226���Wf�FV�6R&F�W"F��FVƗfW&VBv�&���BW&f�&��6R��@���W2�VvV�w2��&�ƗG�F��VWB6���W6RFV��BF�R6�V&W7B66�G�6��7G&��B��BF�7F��wV�6�W0�5��Bw2f�gF��vV�W&F���6����F�G��E$��B�DE#U��72�&�GV7F���6���2g&��FW7FVB��V�B�@��$�&�w&W72��B�G2���VB��B"dBƖ�Rg&��6���W&6���WGWB��BW6W2F�R6��v�RV�F�Ɩ氢CrFV&F�v���ǒ26��R�7V6�f�26����V�BWf�FV�6R�F�R&R�W&��B66V�B66�V�F�f�2Т6��WF��rW"�2&6�w&�V�B&V6W6R�Bv2V&Ɩ2'�F�RW&��BV�C��BF�W2��BV�FW"�Ww5��G0��"W7F&Ɨ6���FV��G&���r�"��fW&V�6R&�G��&V6�V6�VBV�6��vVB��W&bcb�&W�6�F�'��VG0��B7W'&V�B��fW&V�6U�6�fW&vR����F6�VB7W'&V�B6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU�� �ERcrv2fW&�f�VB��B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFR�@�V��FW�VBWf�FV�6R�FV�FW'2���F�f�GV�&V�6��&�'F�f7G2�BV�F�66��6VB&�GV7F����"7W7F��W ��WG&�72�6�W&6U��G3���66V�B�66�V�6R�##b�r�# ��6��R��F�72�##b�����7��B�sR�G&��##b���# ��7��B���B�##b������V�F�Ɩ��Cr�FV&F�v��##b�������VvV��6���V7B�##b���p����fW&V�6W����WF��G&��##b���@����W&b֖�fW&V�6P����W&b�G&���p��FFS�##b���# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�������F�V@��ƖvV�@�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"R�#"�FFVB&WWFW'2r6WFV�&W"# �&W�'BF�BƖ&&V�fV��VBBԆVBw2��V�wRc��B��rF&vWG2##r�72&�GV7F����@�6���W&6��&V�V6R�V&ƖW"F��F�R2##r&�F�&W�'FVB�����&W6W'fVBF�B��FW"FFV@�&�F��B&V6�&FVBF�R�WrF�&VR�F��W2�Ӄ�W&f�&��6R�BS��6��6�W7FW"6���20�&W�'FVBF&vWG2&F�W"F��v�&���BWf�FV�6S�&V6�6������FV���vW"�F����w��&V6�fW'���f'&�6F����B�V��'�FFvW&R��BF�66��6VB�FFVBƖvV�BFV6�����v�W22G&6�VB��F��70�7WƖW"gFW"�G2��BR�cr&��Ɩ�����r���r���v�F����VB���6F���2�b&�WBS"�R�b�W@�&�6VVG2F�f7FW"�F�6��&�GV7B�B6��"dB�B#R�RF�66�G��F�R&V6�&B6W&FW26�F��&�6VB�BfV�F�"�Ɨ7FVBCr�r��eB�'Ff�Ɩ�g&��fW&�f�VB��v��7VVB6���V�B֗����V�B��7W7F��W"VƖf�6F����BF��W7F�26����V�B&�fV��6R�%4�6W"�4����B����6V&6�W2&�GV6V@���&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBF�RV�6��vV@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w0�7W'&V�BF6�&�&C����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr� ��V����\;�v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W"6���&Wf�WrFFW2�6�W&6U��G3���Ɩ&&�c��##b���# ��ƖvV�@��ƖvV�B֗��##b���# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#0�7FGW3�'F���6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VBF�W6RR6���W2v�F�6WFV�&W"b�#2�fW&����BgV��6�W&6R&WfƖFF����`�WfW'�&�f��R�&6�f���VB�VvV�w2&��'�FW����V�B�FW7F��rF�66��7W&R�B��FW&6���V7BMx㝭�G����ƭy��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��&�&V��wV�v�p�����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2��FFVB&�&V�w2�7F�&W"��&��'�f�Ɩ�rF�B�G2V���VB�W�B�vV�W&F���&�GV7B6���WFVB'&��r�W�BV�FW&V@��7B�6�Ɩ6��fƖFF����F�R6���W�V7G27W7F��W"6�Ɩ�r�Bf��V�R&�GV7F����fW"F�P�6�֖�rV'FW'2�7V&�V7BF�6F�6f7F�'�fƖFF����BVƖf�6F���F��2�2F�W&Vf�&R�V�v��VW&��r֖�W7F��R&F�W"F��7W7F��W"F�F����VƖf�6F����"&�GV7F�����6�&V6�&FVB&�&V�w2&W7B�Vff�'G2�6V�V�B�b3֖�Ɩ���6�&W2B��C3��V6��f�"W7F��FV@��WB&�6VVG2�b��CB��r&��Ɩ���F�Rf�Ɩ�r���6FW2sRF�7Wǒ�6���&�7W&V�V�B��&�GV7F���&VF��W72�B6���W&6�Ɨ�F����#RF�&W6V&6��6�gGv&R�BV6�7�7FV�v�&���@�RF�v�&���r6�F��'WBƗ7F��r&�f�&V���VB&W&WV�6�FR�BF�RG&�67F���֖v�@���B&�6VVB�2�b6WFV�&W"3F�R6����BW6VB��C��"�֖�Ɩ����"BR��b�G2&��"�Vǐ��6V�V�B&�6VVG2�F�Rf�Ɩ�rF�B��B��RF�R�Wr&�GV7B�"F�66��6R7V6�f�6F���2��f'&�6F�����V��'����V�B�7W7F��W'2�&V�6��&�2�7WƖW'2�&W6W'fVB66�G��"6���W&6���6��֗F�V�G2���ǒ&�&V�w2&�f��RFFRGf�6VB�7W'&V�BV&Ɩ2&V6�&G2f�"%4�6W"�wV�v�p��B���w6V֒F�B��BW7F&Ɨ6���F�W"VƖf�VB&�GV7F����FW����V�B�7W7F��W"�F�F���� �&V�6��&�֖�W7F��S�6V&6����ǒ6�fW&vRF�B��BGf�6RF�V�"&Wf�WrFFW2���W&bG&���p�cb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���V@�BF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���Wr�F6�VB6���W6P�66V�W&F�"6��&�6��v��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���&�&V���6V�V�B�##b�����&�&V��wV�v�p�����w6V����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b���7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p����WfF �����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"B��FFVBF�R�7F�&W"��F������Wr�VƗG��&�GV7F�fR�f�&6W2wV�F�6R26�����WWG&��Ɩ7�WfV�B��B6��2f�"f7FW ��F�V�'��B6�&R�FV6�����w�FWfV���V�B�7G&��vW"6��WFR��v�&�F���BFF7Wǒ���F������FVw&FVB6��WF��r��vW"�WGv�&��B'&�FW"�F�F���7&�72��GW7G&�W2�@�FWf�6W2�F�R6�RwV�F�6R&WV�&W2��6�ǒ&�&�FR����r�Bv&�2v��7B�W&B�Ɩ�P�6��WF�F����'V&&�W2�&Ɩ�B��fW7F�V�B�BGWƖ6F�fR��6�W��6�����B6WG2�Ɩ7��F�&V7F���'WB&�f�FW2��V�F�f�VB'VFvWB����V�V�FF���66�VGV�R�&�7W&V�V�Bv&B� �6������6F������g&W6�VƖg���r&�GV7F����FW����V�B�7W7F��W"�F�F����f���6��p��"&V�6��&�֖�W7F��Rv2W7F&Ɨ6�VBf�"F�R67&VV�VB6���W2�6���&�f��R&Wf�WrFFP�Gf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���F6�VB6���W6R66V�W&F� �6��&�6��v��7B7W'&V�B'V&���t#3�ԓ3SU��ERcr�"�V����\;�7�7FV�2���F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���6�����Wr�VƗG��&�GV7F�fR�f�&6W2��Ɩ7��##b�������W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b�� �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p����WfF �����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"R�"�&6�f���VB��WfF"w0�6WFV�&W"#��ff�6����FW&��&W�'C�f�'7Bֆ�b&WfV�VRv2$�"�CR�r֖�Ɩ���uuR&WfV�VP�v2$�"�b�֖�Ɩ���BF�R6���&W�'FVB��&RF��3�7W7F��W'2���fW&V�6R�&�GV7@�&WfV�VRv2$�"cSB�"֖�Ɩ���BG&���r�&�GV7B&WfV�VRv2$�"#c�֖�Ɩ���F�Rf�Ɩ�p�F�W2��B��R7W7F��W'2�"F�66��6RV�B6���V�G2�&�GV7B֗��WF�Ɨ�F����"v�&���@�&W7V�G2�F�F�w&�72�&v��fV��F�r�"R��WB&�f�Bv2$�"b�2֖�Ɩ�����w6�FR$� �sc�"֖�Ɩ��f�"�f�VRv����W&F��r66��WFf��rv2$�""�s�b&��Ɩ���B&�'&�v��w0�&V6�VB$�""�c�r&��Ɩ�����6��WF��r�6��WF���2&WfV�VRfV��c��R֖B7Wǒ�&W6�W&6P�6��&F��F����BFVƗfW'��66�VGV�RF�W7F�V�G3�F��2F�W2��B'��G6V�bW7F&Ɨ6�7V6�f�0�6��6��'FvR�"7W7F��W"6�6V��F������ǒ��WfF"w2&�f��RFFRGf�6VB�7W'&V�BV&Ɩ2&V6�&G2f�"F�P��F�W"67&VV�VB6���W2F�B��BW7F&Ɨ6���F�W"g&W6�VƖg���r&�GV7F����FW����V�B��7W7F��W"�F�F����f���6��r�"&V�6��&�֖�W7F��R���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���F6�VB6���W6R66V�W&F� �6��&�6��v��7B7W'&V�B'V&���t#3�ԓ3SU��ERcr�"�V����\;�7�7FV�2���F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����WfF"�##bփ�&W�'@��wV�v�p�����w6V����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b�� �7FGW3�WFFV@�6���W3���&�&V���VvV��7V��'���Т&�F�RvVV�ǒ'&�Vf��r�V�W"f�"F�RgV�ǒ6��6VB�7F�&W"R�W&��B�B&WF��VB���V&ƖW"VF�F���2�VF�FVB��f�W"6���6����W&��BWfV�G2�BF�V�"�&�v���6�W&6W2��W6��rF�R�'6W'fF�'�w2F�ǒ�fW&���rV�vƗ6��B6���W6R6V&6�W27&�72b&��&�G��@���FW7B�&Wf�WvVB6���W2F�6�V6�v2�F�R6���WFRVF�F���6W&FW2&�&V�w2&W�'FV@��7B�6�Ɩ6��'&��r�Wg&��fƖFF����7W7F��W"6�Ɩ�r�VƖf�6F����Bf��V�R&�GV7F��㰢�B�6�6W&FW2&�&V�w2&��6VB��CB��r&��Ɩ���6V�V�Bg&��6���WFVBgV�F��r��&W6W'fVB66�G��B&VƗ�VB�WGWB�F�R�F����wV�F�6R6��&��W27G&��vW"�6��WFR���v�&�F���FF�BF�F���7W�'Bv�F�6��G&��2v��7B'V&&�W2�&Ɩ�B��fW7F�V�B�@�GWƖ6F�fR��6�W��6����'WB&�f�FW2��'VFvWB�66�VGV�R�&�7W&V�V�Bv&B�"6��琢���6F�����VvV��BV�6���w27&�72�Ɩ6V�6R�2&V6�&FVB2v��&��W�6��vR&F�W"F��6���7Wǒw&VV�V�B�"&��b�b��VB66V�BFWV�FV�7����W&bG&���rcb��@���fW&V�6Rcb��VG2&V���VBV�6��vVB��B��fW&V�6U�7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒVF�FVB���FF�F�����FW&�����W&��BWfV�Bv0�W7F&Ɨ6�VB�B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFP�FW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R����V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���&�&V���6V�V�B�##b�����6�����Wr�VƗG��&�GV7F�fR�f�&6W2��Ɩ7��##b������VvV��V�6����FV�B�Ɩ6V�6R�##b��P���VvV��V�6����FV�B�Ɩ6V�6R�&WWFW'2�##b��P����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��0�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p�����w6V���V��V熖��F�V@�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"b�2�BԆVBv2�6�WFFV@�&V6W6RF�R66WFVB�&�WB6�W&6RF�&V7Fǒ6�fW&VB�B�&WWFW'2&W�'FVB�D2f�'7Bֆ�bFF�6��v��r�VvV�6��VBSs"��66V�W&F�'2��6���f�"#�R6�&R�v�F�BԆVBBrR��6�'&�6��BRR�B�V��V熖�BBR�F�F�6���V�G2vW&R"��֖�Ɩ����B����uRV�G2&�6P��V&ǒ�RF��b֖�Ɩ���F�Rf�wW&W27G&V�wF�V�F�F���Wf�FV�6R'WBF���B�FV�F�g��&�GV7B֗��'W�W'2�66WFVBFVƗfW&�W2�WF�Ɨ�F����v�&���G2�&�6W2�"6����V�@�&�fV��6R��BW��'B&W7G&�7F���2��R�&�WB6�&RV�7V�F&�R2g&��F�W"�W&f�&��6P�6��&�6����V��V熖�w2�7F�&W"&��'�&V�V6R�B$Ɩ�bw2&V�V6R�c�B&W�6�F�'�FFV@���7V7F&�R�&6�V�BWf�FV�6R7&�726��v�R��B�V�F����FR��FW2�&�66�VGVƖ�r��44��6���V7F�fW2�B��RĔ$U$�&W&�GV7F����fV�F�"F6��B7��6�&�旦F���&W7V�G2�6���F6�VBf�&V�v�ֆ&Gv&R6��G&���F�66��6VB�vW"�"6�7B���FWV�FV�B&W&�GV7F���� �&�GV7F����7W7F��W"Wf�FV�6R����w6V֒��vV�V�B&W�'FVB�FW&���&FW"w&�wF��&�GV7F���&W77W&R�66�G�W��6����B6��gBF�v&Bg&��B�V�BG&6��B6�V��rWV��V�C����&FW"��WGWB�66WFVB�F����f"���FR�"VƖf�6F���6�V�G2vW&RF�66��6VB��BgV�FV@�W��6���&��V7B&V���2FV��VB���ǒ�VvV��6�'&�6����V��V熖��BԆVB�B���w6V�&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&B7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���6������66V�W&F�"�6�&R�&WWFW'2�##b��0���V��V熖��&Ɩ�b�##b�� ��&Ɩ�bַV��V熖��7W�'@�����w6V֒�66�G��##b�� ��GWb�g@��wV�v�p����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��@�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ���WF������&R�F�&VG0��6��fW'6P�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"r�B�&6�f���VB5��Bw26WFV�&W �#�&�&B�&�fVB&��6�F�W6R$�"2�&��Ɩ���b�fW"�&�6VB��gV�G2F�v&B$�"#B��&��Ɩ��FV6�����w��&W6V&6�&��V7B�B$�"R�&��Ɩ��F�v&B$�"�&��Ɩ��6V6��B��6P�E$�vfW"&6��V�BFW7F��r&6R�F�R&��6�&V���VB7V&�V7BF�6�&V���FW"&�f��@�F�W2��BW7F&Ɨ6�6��7G'V7F������7F��VBFW7B66�G��VƖf�VB�WGWB���V�B�"�$Тf��V�R��T2w2�7F�&W"2f�Ɩ�rFFVB&��6VB$�"3R֖�Ɩ����fW7F�V�B����v�R���w�V�v�F���$�"#S֖�Ɩ��&�V�B��FV�FVB'Fǒf�"��'&�B�&��F��r�BF�֖2�&V�WV��V�@�&W6V&6��B7W7F��W"FVƗfW'��F�Rw&VV�V�B&V���VBV�6�v�VB��B&W�'FVB7W7F��W �fƖFF���F�W2��BW7F&Ɨ6�66WFVB&�GV7F���F���2�F�&�Vv�WB�&VƖ&�ƗG��"F��W7F�0�6����V�B&�fV��6R���ǒ5��B�B�T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2����F�W"g&W6��VƖg���r&�GV7F����FW����V�B�7W7F��W"�F�F����"f���6��r֖�W7F��Rv2W7F&Ɨ6�VB���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6P�cb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&@�7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B� ���7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�V@�6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F�����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���7��B��fW&gV�B�&��V7G2�##b���#����V2���v�R֦��w�V�֖�fW7F�V�B�##b��0����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��P�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p���WF������&R�F�&VG0�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"��R�&6�f���VB6�V�FW'���Bw0���FWV�FV�B6V6��B�V'FW"��B6���V�BW7F��FS���D2�V�BBR'�f��V�R�&涖�rF��&@�&V���B6�7V�r�B4���旂�v�F�6���V�G2W#"R�V"�fW"�V"�BRRV'FW"�fW"V'FW"�6�V�FW'���B&�VB��D2f�gF�'�&WfV�VR&V6W6R�G2֗�&V���VB6��6V�G&FVB����vW"�&�6V@�6��7V�W"Ɩ6F���2v�F���r6�&R�bFF�6V�FW"V�FW'&�6R54G2�6�F�Rf��V�R&涖�p�F�W2��BW7F&Ɨ6�V�FW'&�6R�VFW'6���&�GV7F�����V�B�7W7F��W"VƖf�6F����"7WǒТ6���6�fW&V�v�G���6�FFVBF�R6��v��7F�6�W�6��vRw2VwW7B#&V6�&B66WF��r��D2w0�Ɩ6F���f�"&��6VB$�"32&��Ɩ��5D"�&�WB&�6R�F�R&��V7Bv2V�FW"��V�'���F�R&V6�&BF�W2��BW7F&Ɨ6�6���WFVBf���6��r�f���&�6VVG2���7F��VB66�G���VƖf�VB�WGWB�"gV�ǒF��W7F�2WV��V�B�B�FW&��2f��r���ǒ��D2&V6V�fVB�Wp�&�f��R�&Wf�WrFFR�7W'&V�B�ff�6���"�&�v����6�W&6R6�V6�2f�"F�R�F�W"67&VV�V@�6���W2F�B��BW7F&Ɨ6���F�W"g&W6�VƖg���r&�GV7F����FW����V�B�7W7F��W"ТF�F����"f���6��r֖�W7F��R���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&B7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����F2�7F"��&�WB֗��##b���#��6�V�FW'���Bז�F2�"�6���V�G2�##b��� ����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��`�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p���WF������&R�F�&VG0�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"��b��WF�w2�7F�&W"bf�Ɩ�p�f�&V67B��V'��6WFV�&W"&WfV�VR�b$�""�SC�2�b&��Ɩ���BGG&�'WF&�R�WB&�f�B�`�$�"S�b��sr�"֖�Ɩ���6��&VBv�F�$�"3CR�R֖�Ɩ����72�V"V&ƖW"�F�Rf�&V67@�v2V�VF�FVB��BF�Rf�Ɩ�r6�B��7&V6VB��fW7F�V�B��7Wǒ�6���6���W2�6�6��G&�'WFVBF�F�R&�f�BGW&�&�V�C�7W7F��W"��W2�V�G2�v�&���G2�WF�Ɨ�F�����f'&�6F����B�V��'�&�fV��6RvW&R��BF�66��6VB��U$w27V'6�F�'�6�v�VB�w&VV�V�@�F�7V�&R��'&�B�&��F��rWV��V�BFWfV��W"�V��R���vV�f�"$�"CB�b֖�Ɩ���6��6��p�6��F�F���2&V���VB��B&W�'FVB&�GV7BfƖFF����6���V�G2�B&WVB�&FW'2F���@�W7F&Ɨ6���VB�7W7F��W"66WF�6R�&�GV7F���W6R�F�&�Vv�WB���V�B�WF��R�"F��W7F�0�6����V�B&�fV��6R�6�V�FW'���B&W�'FVBF�B��D2v26��GV7F��r"dB�BFW7F��r�b�6V6��B�vV�W&F���bv"�DE#RFWf�6RB&�Vv�ǒR�vfW'2W"���F�v�F����72�&�GV7F�������G2'F�6�R�6�FW67&�&VBb�RR7W'&V�B��B6���V�B6�&Rv�F��WBV&Ɩ6ǐ�&V6��6�Ɩ�rF�B&6�2v�F�F�RV&ƖW"BR6V6��B�V'FW"W7F��FR�6�&�F�vW&R&WF��V@�v�F�F�RF�67&W�7�F�66��6VB���ǒ�WF���U$�B��D2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6P�cb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&@�7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B� ���7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�V@�6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F�����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3����WF����R����F��f�&V67B�##b��`���W&��V��R֦��vV��7V�6�F����##b��`��6�V�FW'���Bז�F2֗��G&��##b��0����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��p�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p�����&R�F�&VG0��6��fW'6P�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"�r�5��Bw2�7F�&W"rf�Ɩ�p�f�&V67B��V'��6WFV�&W"&WfV�VR�b$�"#cR��#s�&��Ɩ���BGG&�'WF&�R�WB&�f�B�`�$�"3��3"�R&��Ɩ���6��&VBv�F�$�"3"��B&��Ɩ���b&WfV�VR�B$�"R�#�&��Ɩ����72�V"V&ƖW"�F�Rf�wW&W2vW&R&VƖ֖�'��BV�VF�FVB�5��BGG&�'WFVBF�R6��vP�F���v�W"E$�&�6W2���v�WF�Ɨ�F������v�W"�WGWBW"vfW"�66�R�B&�GV7B�֗����&�fV�V�G2�'WBF�B��BF�66��6RV�G2�&�GV7B��WfV�&WfV�VR���V�B�7W7F��W"֗���$Тf��V�R�"F��W7F�2WV��V�B�B�FW&��2&�fV��6R�&6�f���VBF�RwW���V�6��6V�FW"w0�&W�'FVB�W&F��r֖�W7F��S���"��EB3S6&G2���6R��RvW&R�W&F��r�vw&VvFP�FV��Bv2�V"gV����B�B��&RF��#7W7F��W'2vW&RFW7F��r�"W6��rF�R6�W7FW"�F��0��2��VBFW����V�B�BF�F���Wf�FV�6R���Bv�&���B��WfV�WF�Ɨ�F����&VƖ&�ƗG����vW"�"�F6�VBf�&V�v�ֆ&Gv&R&V�6��&��F�R6���VBgV���7F6�F��W7F�27FGW2�6��6�2V&Ɩ26����V�B��WfV�&�fV��6R���ǒ5��B�B���&RF�&VG2&V6V�fVB�Wr&�f��RТ&Wf�WrFFW2���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#��B��fW&V�6Rcb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w0�7W'&V�BF6�&�&BƗ7FVB'V&���t#3�ԓ3SU��ERcr�B�V����\;�7�7FV�2'WB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�B���F�f�GV�&V�6��&�'F�f7G2vW&R��@�W��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���0���6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W �66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���7��B���R����F��f�&V67B�##b��p�����&R�F�&VG2�wW���6V�FW"�##b��� �����&R�F�&VG2�wW���6V�FW"�&��'��##b��`����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�����FW&���6��vP�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"���F�R7vVWf�V�B���Wvǐ�fW&�f�VB6�W&6RW7F&Ɨ6���r�FW&��6��vR��VƖf�VBFVƗfW&�W2�&�GV7F���f��V�R� ���V�B���VB�7W7F��W"66WF�6R��$�&�fV��6R�F���FW����V�B��"7W'&V�B66V�W&F� �W&f�&��6R�6V6��F'�6�fW&vR�b�T2w2F�&VR�F��f�fRזV"WV��V�B�FWfV���V�B��F�@���BW7F&Ɨ6�6���WFVBF���2�7W7F��W"VƖf�6F����"&�GV7F���W6R&W���BF�RW��7F��p�&V6�&B�&��6VB6���W6RFV6�����w��W��'B6��G&��2vW&R��BGG&�'WFVBF�G&6�VB6���w0��W&F��r6��vR���6���&�f��R�&Wf�WrFFRGf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�����FW&���6��vP�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W""���F�R7vVWf�V�B���Wvǐ�fW&�f�VB6�W&6RW7F&Ɨ6���r�FW&��6��vR��VƖf�VBFVƗfW&�W2�&�GV7F���f��V�R� ���V�B���VB�7W7F��W"66WF�6R��$�&�fV��6R�F���FW����V�B��"7W'&V�B66V�W&F� �W&f�&��6R�5��Bw26��&��&��6VB$�"��"�&��Ɩ��6�&R'W�&6�f�"V����VP���6V�F�fW2�"6�F�&VGV7F����'WBF�R&��6�&V���VB7V&�V7BF�&�&B7F����BF�@���Bf���6R�"W7F&Ɨ6�&�GV7F���66�G��VƖf�VB�WGWB�"�$�f��V�R�&��6V@�6���W6RFV6�����w��W��'B6��G&��2vW&R��BGG&�'WFVBF�G&6�VB6���w2�W&F��p�6��vR���6���&�f��R�&Wf�WrFFRGf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�WFFV@�6���W3����V0��6�'&�6���7��@���VvV����V��V熖���WF����W&��F�V@����F0�7V��'���Т&�F�RvVV�ǒ'&�Vf��r�V�W"f�"F�RgV�ǒ6��6VB�7F�&W""��W&��B�B&WF��VB���V&ƖW"VF�F���2�VF�FVB��6��6���6�F�66��7W&W2V&Ɨ6�VB��F�RW&��B�BF�V� ��&�v���6�W&6W2�W6��rF�R�'6W'fF�'�w2F�ǒ�fW&���rV�vƗ6��B6���W6R6V&6�W0�7&�72"&��&�G���Bf�W"��FW7B�&Wf�WvVB&�f��W2�W2f���7&�72�6V7F�"v6V&6��&WWFW'2�&W�'FVB�D26���V�BFF7G&V�wF�V�VBWf�FV�6R�bF��W7F�266V�W&F�"F�F���'W@�F�B��BW7F&Ɨ6�&�GV7B֗��66WFVBFVƗfW&�W2�WF�Ɨ�F����"�F6�VBW&f�&��6R�5��@��B�WF�7WƖVB&VƖ֖�'�V�VF�FVBWf�FV�6R�b6���W&6��66�Rv�F��WBF�RV�B��7W7F��W"���V�B�v�&���B��$��f��V�R�"&�fV��6RFWF���VVFVBf�"&�GV7F���6��6�W6���2��U$w26�v�VB7V�6�F����B�T2w2&��6VB��fW7F�V�BF&vWFVB��'&�B�&��F��rWV��V�B��'WB6��6��r�&�GV7F���W6R�B�W&F��rWf�FV�6R&V���VB��6���WFR�6�V�FW'���BFW67&�&V@���D2w2E$�v�&�2&W6V&6��BFW7F��rv�F��WB�72�&�GV7F�����f�"F�RFW7FVBFWf�6R���W&bG&���rcb��B��fW&V�6Rcb��VG2&V���VBV�6��vVB��B��fW&V�6U�7WƖVB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒVF�FVB���FF�F�����FW&������W&��BWfV�Bv2W7F&Ɨ6�VB�B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf� �&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3����V2���v�R֦��w�V�֖�fW7F�V�B�##b��0��6������66V�W&F�"�6�&R�&WWFW'2�##b��0��6�V�FW'���Bז�F2֗��G&��##b��0��7��B���R����F��f�&V67B�##b��p���WF����R����F��f�&V67B�##b��`���W&��V��R֦��vV��7V�6�F����##b��`����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2�#�F�V�&�F&vWFVB&6�f����6V&6�W27&�72�V��'��66V�W&F�'2�BWV��V�B�FFVB5��Bw2�7F�&W"���V�6V�V�B�b�6V�b�FWfV��VBDc"fW'F�6��7W'&�V�F��r�vFR6V��&6��FV7GW&R�B���VB6W'fW"DE#R$D��Т'��V"�V�B##b�F��2�2&�F����B6���WFVB&�GV7B�VƖf�6F����"f��V�R�&�GV7F���֖�W7F��S�7VVB�66�G����FR�7W7F��W'2�B��FV�FVBf��V�RvW&RV�F�66��6VB�&6�f���V@��7F�&W""W���&�F���&W�'F��r��V�f��Rw2U4�cB�7WW&��FRv�F��'F��v���V�V7G&�6��@��F�6�66�R��WB��FW&6���V7G2�v�F��WBG&VF��rF�7��2��W&F��rFW����V�B� �W&f�&��6R&W7V�B�F�R6�R&W�'BFFVB�T2w26���VB66�R�b��&RF���Ã7V�V�F�fP�&V7F���6��&W'2���fW"##&�GV7F���Ɩ�W2�B�G2�Wvǒ�V�6�VB��v��7V7B�&F��WF6���B3����ǖ&FV�V��B7�7FV�2�7W7F��W'2�&�GV7B֗����FW2�WF�Ɨ�F����WF��R���V�@�6��G&�'WF����BVƖf�6F����bF�R�WrF���2&V���V�F�66��6VB���ǒ5��B�V�f��R�@��T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����V�����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���7��B�Fc"�&F����##b����wV�����W���##b�� ����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W� → —
- schema_version: — → 2
- baseline_date: — → 2026-09-11
- last_research_at: — → 2026-10-20
- last_automated_run_at: — → 2026-10-20
- last_run_status: — → updated
- coverage: — → Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, plus targeted backfill searches. Added CXMT's reported 4F2 server-memory roadmap, Enflame's displayed ESL64-0 supernode and AMEC's reported installed reaction-chamber scale and two memory-equipment launches. These disclosures establish announced architecture, integration and commercial-scale evidence but not completed 4F2 products, operating supernode deployments, new-tool customer qualification, workload performance, yield or provenance. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 and Inference v6.1 repository heads were unchanged, and InferenceX still supplied no matched current Chinese accelerator result. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.
- timezone: — → Europe/London
- runs: — → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 13-20, then ran targeted backfill searches across memory, accelerators and equipment. Added CXMT's October 10 announcement of a self-developed 4F2 vertical-surrounding-gate cell architecture and a planned server DDR5 RDIMM by year-end 2026. This is a roadmap, not a completed product, qualification or volume-production milestone; speed, capacity, node, customers and intended volume were undisclosed. Backfilled October 12 exhibition reporting on Enflame's ESL64-0 supernode with orthogonal electrical and optical scale-out interconnects, without treating a display as an operating deployment or performance result. The same report added AMEC's claimed scale of more than 8,800 cumulative reaction chambers in over 220 production lines and its newly launched high-aspect-ratio etch and 300 mm molybdenum ALD systems. Customers, product mix, nodes, utilization, uptime, yield contribution and qualification of the new tools remain undisclosed. Only CXMT, Enflame and AMEC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-4f2-rdimm-2026-10-11","guoxin-expo-2026-10-12","mlperf-training","mlperf-inference-6-1","inferencex"]}]
2026-10-2095c62cf ↗
research: add October 20 evidence updates
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 19 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads and submitter listings plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 20 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads and submitter listings plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-19 → 2026-10-20
- Updated: AMEC
5 field changes
- assessment: AMEC's half-year report provides production-line, customer-validation and repeat-order evidence across etch and tungsten deposition. It does not name customers or disclose advanced-node yield, uptime, throughput, installed capacity or complete-process independence. Its proposed investment in Qinghe Jingyuan supports an advanced-packaging-equipment supplier but is not evidence that bonding tools have completed customer qualification or entered volume production. → AMEC's half-year report provides production-line, customer-validation and repeat-order evidence across etch and tungsten deposition. It does not name customers or disclose advanced-node yield, uptime, throughput, installed capacity or complete-process independence. Its proposed investment in Qinghe Jingyuan supports an advanced-packaging-equipment supplier but is not evidence that bonding tools have completed customer qualification or entered volume production. Reported product and installed-base figures from the 2026 Guoxin Expo strengthen scale evidence but do not identify production nodes, tool utilization, acceptance criteria or customer economics.
- facts: [{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]}] → [{"label":"Reported installed reaction-chamber scale","value":"China Star Market reported that AMEC had more than 8,800 cumulative reaction chambers operating in over 220 domestic and overseas production lines at more than 100 customers by June 2026. The report did not name customers or disclose product mix, nodes, utilization, uptime, yield contribution or repeat-order economics.","evidence":"reported","as_of":"2026-06-30","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Advanced-memory tool launches","value":"AMEC displayed its newly launched Primo XD-RIE 70-to-90:1 aspect-ratio etch system and Prefima Unicore AM 300 mm molybdenum ALD system. Announced capabilities were not accompanied by named-customer qualification, accepted deliveries, production use or operating metrics.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"Reported foreign-fab evaluation","value":"Reuters reported Samsung and SK Hynix had evaluated AMEC etch tools at their Chinese fabs, while both companies denied testing AMEC equipment for use in China and no deployment decision was reported.","evidence":"reported","as_of":"2026-08-05","source_ids":["amec-evaluation-reuters-2026-08-05"]},{"label":"Advanced etch production","value":"Company reports Primo Nanova LUX-Cryo in stable volume production on a customer's next-generation product line, while Primo Angnova remained in customer certification with multiple repeat orders.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"Tungsten deposition qualification","value":"Company reports its tungsten CVD, high-aspect-ratio CVD and ALD series passed onsite validation at key memory and logic customers and received repeat production orders; customers and production metrics were not disclosed.","evidence":"vendor","as_of":"2026-06-30","source_ids":["amec-2026-h1-report"]},{"label":"CSEAC 2026 tool launches","value":"Six announced tools span very-high-aspect-ratio etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; customer qualification remains unverified.","evidence":"reported","as_of":"2026-09-01","source_ids":["amec-cseac-2026-09-05"]},{"label":"Advanced bonding-equipment investment","value":"AMEC approved up to RMB 35 million of additional investment in Qinghe Jingyuan as part of a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment; the agreement remained unsigned and production qualification was undisclosed.","evidence":"primary","as_of":"2026-10-09","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13"]}]
- watch_for: ["Advanced-process customer qualifications","Repeat orders and production utilization","Signed Qinghe Jingyuan financing and independently verified bonding-tool customer acceptance"] → ["Advanced-process customer qualifications","Repeat orders and production utilization","Signed Qinghe Jingyuan financing and independently verified bonding-tool customer acceptance","Customer qualification and production metrics for XD-RIE and Unicore AM"]
- source_ids: ["amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13"] → ["amec","amec-evaluation-reuters-2026-08-05","amec-2026-h1-report","amec-cseac-2026-09-05","amec-qinghe-jingyuan-investment-2026-10-13","guoxin-expo-2026-10-12"]
- reviewed_at: 2026-10-14 → 2026-10-20
- Updated: CXMT
5 field changes
- assessment: Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. An unaudited nine-month forecast reports a sharp revenue increase and profit turnaround amid higher DRAM prices and utilization, strengthening commercial-scale evidence without disclosing unit volume, yield or product mix. → Fifth-generation commodity DRAM and LPDDR6 are company-reported in mass production, with Xiaomi providing a named LPDDR6 handset design-in. Adequate domestic HBM volume and frontier HBM performance remain separate milestones. Proposed post-IPO research and back-end-testing projects would expand vertical integration, but approved funding is not completed capacity. The reported NAND project is an R&D-line plan, not evidence of qualified commercial output. An unaudited nine-month forecast reports a sharp revenue increase and profit turnaround amid higher DRAM prices and utilization, strengthening commercial-scale evidence without disclosing unit volume, yield or product mix. Its announced 4F2 VSG roadmap targets a server DDR5 RDIMM by year-end 2026, but no completed product, qualification, capacity, speed or density was disclosed.
- facts: [{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}] → [{"label":"4F2 server-memory roadmap","value":"CXMT said its first product using a self-developed vertical-surrounding-gate 4F2 cell architecture would be a DDR5 RDIMM expected by year-end 2026. The company described related word-line, bit-line and junction-process work, but disclosed no completed product, speed, capacity, node, customer qualification or intended volume.","evidence":"reported","as_of":"2026-10-10","source_ids":["cxmt-4f2-rdimm-2026-10-11"]},{"label":"Nine-month earnings forecast","value":"CXMT forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited.","evidence":"primary","as_of":"2026-09-30","source_ids":["cxmt-nine-month-forecast-2026-10-17"]},{"label":"LPDDR6 commercial design-in","value":"CXMT and Xiaomi said CXMT's LPDDR6 entered mass production for the Xiaomi 18 Fold. CXMT reports a 16 Gb die, 16 GB package and up to 12,800 Mbps; bandwidth and power comparisons are internal tests, while volume, yield, reliability at scale and fabrication provenance are undisclosed.","evidence":"reported","as_of":"2026-08-29","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29"]},{"label":"Capacity expansion","value":"Reuters reported two new plants under construction and discussions for a third, with projects projected to lift capacity above 600,000 wafers per month; completion, qualification and realized output remain unverified.","evidence":"reported","as_of":"2026-07-24","source_ids":["china-memory-expansion-reuters-2026-07-24"]},{"label":"Fifth-generation DRAM platform","value":"CXMT says its fifth-generation platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch; independent yield and reliability data were not disclosed.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"LPDDR5X production","value":"Two 24 Gb LPDDR5X products are company-reported in mass production, with at least 50% more gross dies per wafer than the fourth-generation platform on an 8 Gb baseline; gross dies are potential dies before defective units are excluded, not tested yield.","evidence":"reported","as_of":"2026-09-20","source_ids":["cxmt-g5-dram-2026-09-20"]},{"label":"NAND expansion plan","value":"Reuters reports CXMT plans a Beijing NAND R&D production line and has discussed prospective AI-storage customers; no operating date or commercial-production commitment was established.","evidence":"reported","as_of":"2026-09-18","source_ids":["cxmt-nand-2026-09-18"]},{"label":"Proposed research and testing investment","value":"CXMT's board proposed using RMB 13.0 billion of over-raised IPO funds for a technology- research project and RMB 5.0 billion for a second-phase DRAM wafer back-end testing base; shareholder approval, construction, installed capacity and qualified output remained pending.","evidence":"primary","as_of":"2026-09-28","source_ids":["cxmt-overfund-projects-2026-09-29"]}]
- watch_for: ["Customer-qualified HBM shipments","Stack yield and volume","Accelerators shipping with verified domestic DRAM","Independent G5 yield, reliability and customer qualification","Named domestic-equipment contributions to the G5 process flow","NAND trial-line operation and transition to qualified volume production","LPDDR6 shipment volume, yield and additional customer qualification","Shareholder approval, construction and operating milestones for the research and testing projects"] → ["Customer-qualified HBM shipments","Stack yield and volume","Accelerators shipping with verified domestic DRAM","Independent G5 yield, reliability and customer qualification","Named domestic-equipment contributions to the G5 process flow","NAND trial-line operation and transition to qualified volume production","LPDDR6 shipment volume, yield and additional customer qualification","Shareholder approval, construction and operating milestones for the research and testing projects","Launch, independent validation and server qualification of the planned 4F2 DDR5 RDIMM"]
- source_ids: ["cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17"] → ["cxmt-hbm","cxmt-hbm-reuters-2026-08-31","cxmt","china-memory-expansion-reuters-2026-07-24","cxmt-nand-2026-09-18","cxmt-g5-dram-2026-09-20","cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","cxmt-overfund-projects-2026-09-29","cxmt-nine-month-forecast-2026-10-17","cxmt-4f2-rdimm-2026-10-11"]
- reviewed_at: 2026-10-17 → 2026-10-20
- Updated: Enflame
5 field changes
- assessment: TechInsights found TSMC N6 in the examined S60 compute chiplet; this does not identify the provenance of every later product. → TechInsights found TSMC N6 in the examined S60 compute chiplet; this does not identify the provenance of every later product. Enflame displayed an ESL64-0 supernode with an orthogonal, cable-free electrical interconnect and optical scale-out, but supplied no card count, operating deployment, workload, reliability, power or matched performance result.
- facts: [{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}] → [{"label":"ESL64-0 supernode","value":"Enflame displayed an ESL64-0 supernode using its own AI accelerators, an OEX orthogonal electrical interconnect, a cable-free internal design and optical scale-out. The exhibition report did not establish an operating deployment, card count, topology, workload performance, utilization, reliability or power.","evidence":"reported","as_of":"2026-10-12","source_ids":["guoxin-expo-2026-10-12"]},{"label":"IPO-funded development plans","value":"Reuters reports IPO proceeds are intended mainly for fifth- and sixth-generation AI chips, software and large computing systems; funding plans are not delivered products.","evidence":"reported","as_of":"2026-09-11","source_ids":["enflame-ipo"]},{"label":"S60 compute chiplet","value":"TSMC N6, identified by independent teardown","evidence":"measured","as_of":"2026-02-03","source_ids":["enflame-teardown"]}]
- watch_for: ["Adoption beyond Tencent","Manufacturing route for new generations","Independent training and inference results"] → ["Adoption beyond Tencent","Manufacturing route for new generations","Independent training and inference results","Operating ESL64-0 deployments with card count, topology, utilization, reliability and power data"]
- source_ids: ["enflame-ipo","enflame-teardown"] → ["enflame-ipo","enflame-teardown","guoxin-expo-2026-10-12"]
- reviewed_at: 2026-09-12 → 2026-10-20
- Added: AMEC displays high-aspect-ratio etch and molybdenum ALD systems
- Added: CXMT targets a 4F2 server DDR5 RDIMM by year-end 2026
- Added: Enflame displays its ESL64-0 supernode
- Updated: data/research-state.yaml
9 field changes
- schema_version: 2 → —
- baseline_date: 2026-09-11 → —
- last_research_at: 2026-10-19 → —
- last_automated_run_at: 2026-10-19 → —
- last_run_status: updated → —
- coverage: Published the complete October 12-18 weekly briefing after auditing six canonical disclosures and their original sources, using the daily overlapping English and Chinese reviews of 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. The report distinguishes IDC shipment share from performance evidence, preliminary CXMT and MetaX financial forecasts from audited operating detail, hybrid-bonding transactions from production qualification, and YMTC's DRAM research wafers from commercial output. No additional material in-period event was established. MLPerf Training v6.0 and Inference v6.1 heads were unchanged, while InferenceX still supplied no matched current Chinese accelerator result. No dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance. → —
- timezone: Europe/London → —
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}] → —
- : — → Y��x-���jם��i��+��j[h��ܢ��ߎ9�:-jZ.����)W66�V��fW'6��� �&6VƖ�U�FFS�##b�����7E�&W6V&6��C�##b��# ��7E�WF��FVE�'V��C�##b��# ��7E�'V��7FGW3�WFFV@�6�fW&vS��Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2�#��W2F&vWFVB&6�f���6V&6�W2�FFVB5��Bw2&W�'FVBDc"6W'fW"��V��'�&�F��V�f��Rw2F�7��VBU4�cB�7WW&��FR�B�T2w0�&W�'FVB��7F��VB&V7F����6��&W"66�R�BGv��V��'��WV��V�B�V�6�W2�F�W6RF�66��7W&W0�W7F&Ɨ6����V�6VB&6��FV7GW&R���FVw&F����B6���W&6���66�RWf�FV�6R'WB��B6���WFV@�Dc"&�GV7G2��W&F��r7WW&��FRFW����V�G2��Wr�F���7W7F��W"VƖf�6F����v�&���@�W&f�&��6R���V�B�"&�fV��6R���ǒ5��B�V�f��R�B�T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb��B��fW&V�6Rcb�&W�6�F�'��VG2vW&RV�6��vVB��B��fW&V�6U�7F����7WƖVB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�B���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2���F�f�GV�&V�6��&�'F�f7G2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B��WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�F��W���S�WW&�R����F��'V�3���FFS�##b��� �7FGW3�'F���6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��6Т�&�&V��6��� �7V��'���ТF�W6RR6���W2vW&R67&VV�VB���B��gV�ǒ&WfƖFFVB�&VB�&�v���4��R&W�'F��p��B�F�6�7WƖW'2rV&Ɩ26�FW3�FFVB&W�'FVB6���V�B&�w&W76���f�"����Ɩv�@��BV�F�Ɩ��&VB&�&V�w2FFVB&��'����V�6V�V�G3�FFVB6�gGv&R�fƖFF����@�&W6V&6���&�&F�'�WfV�G2�&W6W'f��rF�RF�7F��7F���&WGvVV�V&Ɩ6F����BWfV�BFFW2�&Wf�WvVBF�R��FW�VB�&�v���6WFV�&W"&WWFW'2V�f��R&W�'B�BFFVB���VBW6P��b��&�6VVG2v�F��WBGWƖ6F��r�G2W��7F��r��WfV�B���ǒF�W6Rf�W"6���&Wf�Wp�FFW2Gf�6VB�&��&�G��6���6V&6�W2�6�&Wf�6�FVBF�R6WFV�&W"�$�&�6��p�&W�'B�B�VǒEUb&W�'F��s���FF�F����6���W&f�&��6R�"VƖf�VB�F���6���v0�66WFVB��T2w2�ff�6��6�FR�BF�RgV��f���6��F��W2EUb'F�6�R6�V�B��B&R&WG&�WfVB�f����rW�T2w26���F������V�6V�V�B�F�REUbFW7F��r6�&�����w��&W�'FVB�F�6����GV�P�W��'B&W7G&�7F���2��B�V��'��WV��V�Bf�Ɩ�w3�F���B��fW"VƖf�6F���g&���VFƖ�W2�F�v���FVB�Bf�7V�ǒ6�V6�VB&�&V�w2&��'���bזV"��f�w&��3�FFVB�G2&WfV�VR����72�"dB�BV���VB�7W7F��W"FVƗfW'�6���2v�F�fV�F�"GG&�'WF������B���FWV�FV�BVF�B�4��B4���6V&6�W2F�B��BW7F&Ɨ6�fW&�f�VB�WrVƖg���r֖�W7F��S�F�V�"&�f��P�FFW2&V���V�6��vVB�6�V6�VB��W&bG&���rcb�&W�6�F�'�Ɨ7F��w2���fW&V�6Rcb� �6��6VB�7V&֗GFW"Ɨ7F��w2f�v�D�V"��B��fW&V�6U�w2V&Ɩ26�fW&vR����F6�VB7W'&V�@�6����fW'7W2�g&��F�W"&W7V�Bv2fW&�f�VC���F�f�GV�&W7V�B'F�f7G2vW&R��BW��W7F�fVǐ�VF�FVB���7F�&�6��Ww2�W&��B'&�Vf��w2�f�&V67G2�B66V�W&F�"7V6�f�6F���2&R&WF��VB�&W6V&6�66�R�2'F����FWV�FV�Fǒ�bv�D�V"7��2�"6��VFf�&RV&Ɩ6F���7FGW2�6�W&6U��G3���6��R��F�72�##b�����&�&V��##bփ��&�&V��v���##b���p��&�&V��6��v����"�##b��� ��V�f��R֗���&��&WWFW'0��GWb�&WWFW'0����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���0�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��6Т�6��� ��V��&V�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�RF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W2�FFVB�F�&6�f�V�FF���&��'�Wf�FV�6RF�@�6�'&�6������VB2�F��V��V�&W"v�F�v�fW&��r&�&B�BFV6��6�Gf�6�'�6�V�6���&W&W6V�FF����B&W�'FVBW7G&V�v�&�7&�726����R�'V�F��R�BF�7G&�'WFVB&V2�FFVB�VvV�w2W��7F��r�F��V��V�&W'6���B6��6��"&��R��F�Rf�V�FF���w266V�W&F� ���FVw&F���v�&���rw&�W�F�W6R&RV6�7�7FV��Bv�fW&��6R6�v��2���BW&f�&��6R&W7V�G2�6��6VBF�R&��"�T2f����r�WW6��rFWF��VB66�V�6R�BFV6�����w�F�ǒ&W�'F��r��6���54T2�V�6�W27&�72WF6��T5dB��B��WF��f���FW�6�F����B6�Ɩ6���6&&�FRW�F�����WB7V6�f�6F���2F�7F��7Bg&��7W7F��W"VƖf�6F����Bf��V�R�&�GV7F���Wf�FV�6R�&Wf�WvVB7W'&V�B�ff�6���'Ff�Ɩ��B�Ww2vW2f�"4�&W6V&6�6��v���4����@�V��&V����WrVƖg���r&�GV7F����7W7F��W"�"6���WFR�f��r֖�W7F��Rv2fW&�f�VB��6�F�V�"f7G2vW&RV�6��vVBv���RF�V�"7GV�&Wf�WrFFW2Gf�6VB�&V6�V6�VBF�P�f���6��F��W2EUb7F�'���$�&�6��r6�fW&vR�F�R6WFV�&W"�F�72&W�'B�BV�f��Rw0���f����r�WGW&��rF�R�fW&�v��F�s���FF�F����6���v266WFVB�&Wf�WvVBF�Rd42w0�f���6����V�B�'V�R6�fW&vR'WBF�B��BGF6��V�F�G��7V6�f�2&W7G&�7F���F�����Ɩv�@��"V�F�Ɩ�&V6W6RF�R'V�RF�B��B��RF�V����W&bG&���r�B��fW&V�6Rcb�&W�6�F�'���VG2vW&RV�6��vVB��BF�R6��6VB��fW&V�6R7V&֗GFW"Ɨ7B7F���6��vVB���F6�VB7W'&V�@�6���W6R66V�W&F�"���fW&V�6U�6�fW&vR7F���F�B��B7Wǒ�WV�f�V�B6���W6R&W7V�B�����FWV�FV�B�FW7B�fW'7W2��FW7B&�G�6���v2FFVB���7F�&�6��Ww2�&W�'G2��f�&V67G2�Wf�FV�6R�&V�2�B&��"'V�&V6�&G2vW&R&W6W'fVB�6�W&6U��G3����F�&6��6�'&�6���##b���p���F�&6��6����##b�������V2�76V2�##b���P��6Т�6��� ��V��&V��GWb�g@���&��&WWFW'0��6��R��F�72�##b�����V�f��R֗����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���@�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��wV�v�p���60���vG6��p�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F�6WFV�&W"~( 3B�fW&���rV�vƗ6��B6���W6R6V&6�W2�FFVB�42&��'�Wf�FV�6P�f�"�G2&W�'FVB������w6�RDD2FW����V�B�r�Ɩ��B�D42'F�6�F����U����GV�P�7W�'B�B�V��F�6�֖�FW&6���V7BF�&V7F����F�RF�66��7W&RF�B��B&�f�FR�F6�V@���f��&�Bv�&���B&V�6��&��6�W7FW"F����w��7v�F6��6�Ɩ6��&�fV��6R�"��FWV�FV�@�7W7F��W"FF�FFVB&W�'F��r6�W&6VBF��vG6��rw26������V�6V�V�BF�B�G0��7FW"�SU��V���WfV�4�7�7FV�v26��VBF��Gf�6VB�6�v��r&�GV7F���Ɩ�S���WB6���V�BF�7F��7Bg&��7W7F��W"66WF�6R���V�B�WF��R�F�&�Vv�WB�Bf��V�P�Wf�FV�6R�&V�V�VBwV�v�rw2FV�B&V6�&B�BF�R7W'&V�B�42�B�vG6��r6���vW3��wV�v�rv��VB���Wr&�GV7F���Wf�FV�6R�6V&6����ǒ6�fW&vR�bF�R�F�W"&��&�G��&�f��W2F�B��BGf�6RF�V�"&Wf�WrFFW2�&V6�V6�VB��W&bG&���r�B��fW&V�6Rcb� �&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR�F�P���W&b&W�6�F�&�W2�B���WvW"&W7V�BW6��B��7W'&V�B6���W6R66V�W&F�"7V&֗GFW �v2f�V�C���fW&V�6U�Ɨ7FVB7W'&V�Bt#3�ԓ3SU��BERcr7�7FV�2'WB��WV�f�V�@�7W'&V�B6���W6R�&Gv&R&W7V�B����F6�VB�FW7B�fW'7W2��FW7B&�G�6���v2FFVB���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�Wf�FV�6R�&V�2�B&��"'V�&V6�&G2vW&P�&W6W'fVB�6�W&6U��G3����62��G��##b���@���vG6��r��V��6��##b������wV�v�p���&��&WWFW'0��6��R��F�72�##b�����V�f��R֗����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���@�7FGW3�WFFV@�6���W3����6�V�v����V0��&�&V��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV����vG6��p����WfF ������Ɩv�@���WF����W&��'6�6W ��6֖0����F0���VƖ�w6�V�p�������7V��'���Т&�F�RvVV�ǒ�V�W"f�"F�R&Wf��W2gV�ǒ6��6VB���F��7V�F�W&��B�B&WF��VBF�P�6WFV�&W"W&��B�F��FFRVF�F���V�6��vVB�VF�FVBF�RFV�6V�V7FVB���W&��BWfV�G2�@�&��fW&���rV�vƗ6��B6���W6Rv6V&6�W27&�7266V�W&F�'2�6�gGv&R��V��'����F�6���FW&6���V7G2�ƗF��w&���B6�v��rWV��V�B�FFVB&�&V�w26WFV�&W"rfV�F� �66�V�B�b�VwW7B#�V6�7�7FV��VWGW26���6�6�W&6R�BWfV�C��B6�2F�&6�5U�W6W2�F�&6�&�fFUW6S�B7W�'G26������2�6���V7F�fW2�Be4E"�w&�6GW&R�@�&�f�Ɩ�r���V&Ɩ26�FR&W�6�F�'�v2�FV�F�f�VB�6�F�R&V6�&BF�W2��B��fW"�W&F� �6�fW&vR����F��&�ƗG��W7G&V�66WF�6R�W&f�&��6R�"5TD&�G��F�R6���WFP�'&�Vf��r&W6W'fW2����7F�&�6�&W�'G2�B7FFW2F�B��FFVBf�&V67B��fVB��BW6W0�F�R7W'&V�B6WFV�&W"B&V�6��&�&Wf�Ws����F6�VB6��FV��&'�6���W6R&W7V�Bv��7@�t#3�ԓ3SU��"ERcrv2fW&�f�VB����W&bcb��"��fW&V�6U�6�fW&vR�&V6W6R7�7FV�F�0�6�fW&vR&Vv�6WFV�&W"�F�R6WFV�&W"r�2&V6��7G'V7F���&V���2��6���WFRf�"&�fFP�FW����V�G2�V��FW�VBF�66��7W&W2�FV�FW'2�B��F�f�GV�&V�6��&�'F�f7G2�6V&6����ǐ�6�fW&vRF�B��BGf�6R�F�W"6���&Wf�WrFFW2�6�W&6U��G3���&�&V��F�&6�7W�##b���p��&�&V��##bփ��&�&V��v���##b���p��&�&V��6��v����"�##b��� ���F�&6��6�'&�6���##b���p���F�&6��6����##b�������&��&WWFW'0��6��R��F�72�##b�����V�f��R֗���vG6��r��V��6��##b������GWb�g@����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���P�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p����WfF ���6W@�����w6V�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"��R����WrVƖg���p����v��F�r66V�W&F�"��V��'���F�6�֖�FW&6���V7B�"6V֖6��GV7F�"�WV��V�BWfV�Bv0�fW&�f�VB�f�Ɩ�r�v&Wf�Wrf�V�BF�&VR�FW&��VwW7BF�66��7W&W2F�BvW&R֗76��rg&�ТF�R6���6�&V6�&B��T2w26���&W�'B6�2F�&VR�5WF6�f֖ƖW2vW&R��f��V�P�&�GV7F�����Ɩ�W2B��&RF��S7W7F��W'2�v�F�7F&�R&�GV7F���W6R�"&WVB�&FW'2f� �FF�F����Gf�6VBWF6�7�7FV�2�B&WVB&�GV7F����&FW'2gFW"��6�FRfƖFF����`�GV�w7FV�FW�6�F���7�7FV�2B�W��V��'��B��v�27W7F��W'2��U$&W�'G2��&RF��� �7V�V�F�fRFVƗfW&�W2V6�f�"WF6��dB�BfW'F�6�gW&�6W2�$�"#�c&��Ɩ���bf�'7Bֆ�`�&WfV�VR��B&����WB�b��'&�B�&��F��r�B��v��V�BE5bV�V7G&��F��rF���2��4UB&W�'G0��Dd��f��V�R&�GV7F����7W7F��W"fƖFF����b6�Ɩ6�����F��72V�v��R6��W2��BC�BP��V"��fW"זV"w&�wF���6��WF��r�V�V7G&��72&WfV�VR�V6�6���&V���2fV�F�"Wf�FV�6S��V���VB7W7F��W'2�V�7V6�f�VB&�6W72��FW2��BV�F�66��6VB��V�B�66�G��WF��R��F�&�Vv�WB�"&�GV7F���f��V�RƖ֗BF�R6��6�W6���2�&Wf�WvVB��WfF"�B���w6V֒w0�7W'&V�B�ff�6��vW2v�F��WB66WF��r�Wr&�GV7F����"&V�6��&�6����&V6�V6�V@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vS�����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU��"ERcr�6�727�7FV�0�v2f�V�B�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R&Wf�WrFFW2���7F�&�6�&W�'G2��6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVC�F�RVwW7BWfV�G2&RW�Ɩ6�Fǐ��FR�FFVB&F�W"F��FFVB26WFV�&W"WfV�G2�6�W&6U��G3����V2�##bփ�&W�'@���W&�##bփ�&W�'@���6WB�##bփ�&W�'@����WfF �����w6V����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���`�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p���V��V熖���WF������&R�F�&VG0�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"��b�FFVB���&RF�&VG2p�6WFV�&W"�&��'����V�6V�V�BF�B�B6��VB��2��6&BgV���gV�7F���uR6�W7FW"�F�R&V6�&BG&VG2F��22��&F�W"F��FVƗfW'�Wf�FV�6R&V6W6R��66�VGV�R�uR��FV���6���WFVB6&B6�V�B�F����w����FW&6���V7B��vW"�&V6�fW'��"v�&���B&W7V�Bv2F�66��6VB�&6�f���VB���&RF�&VG2rVwW7B��bזV"&W�'C�6����&W�'FVB&WfV�VRv2$�"�s3`�&��Ɩ���GG&�'WF&�R�WB��72v2$�"�Sb֖�Ɩ����B$�"cc֖�Ɩ���TR6�W26��G&7@�v2gV�ǒFVƗfW&VB�B&V6�v旦VB��G2FV�6R�B��R�eR�66Ɩ�r�Vff�6�V�7��BVffV7F�fRТG&���r�F��Rf�wW&W2&V���fV�F�"�V7W&V�V�G2v�F��WB�F6�VBv�&���B��vW"�"7W'&V�@�f�&V�v�ֆ&Gv&R6��&�6���FFVB�V��V熖�w2F�v���F&�Rf�t�2֖�5�R�$"FW����V�B��vP��B�WF�w2E$�Fv�&Wf�WrF��FFF����W2&W�'FVB�F�&6��W&F�"6�fW&vR0�6��F�&�ƗG�Wf�FV�6R���BW&f�&��6R�"5TD&�G��&V6�V6�VB��W&bG&���r�B��fW&V�6P�cb�&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w27W'&V�B6�fW&vR����F6�V@�6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R&Wf�WrFFW2���7F�&�6�&W�'G2�6�'&V7F���2��f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVC�F�R��bזV"f�Ɩ�r�2W�Ɩ6�Fǒ�FR�FFVB�6�W&6U��G3������&R�F�&VG2�##bփ�&W�'@�����&R�F�&VG2֦B�6��VB�##b�������V��V熖��֖�7��##b��� ���WF��G&��##b���@���F�&VG0���V��V熖����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���p�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�6��fW'6P�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"�r�FFVB&WWFW'2p�6WFV�&W"r66�V�B�b�VvV�w2�Wr�V�6�F&vWG2f�"F�R�cEB��##r�B66V�@��c"��2##r��W2�VvV�w26���F�B�B�B6��VB��&RF���7WW&��FW2F���&P�F��3s7W7F��W'2�F�R&V6�&B&W6W'fW2F�R��FW"FFVB&�F��BG&VG2F�R�WrFFW20�W�V7FVB֖�W7F��W2���BFVƗfW'�Wf�FV�6R��VvV�F�B��B�FV�F�g�7W7F��W'2�7FFRF�P��V�&W"�b66V�W&F�'2W"6��VB7�7FV��"F�66��6R66WF�6R�WF�Ɨ�F����v�&���B���vW"�&VƖ&�ƗG��"��FW&6���V7B�W&f�&��6R&W7V�G2�FFVB�VvV�w26WFV�&W"66�V�@��b6���V�6��FW�����r66V�B6��WFR�B3�B��R7WW&��FR�BFF��r66V�B�Sf� �V�6��6��VC�F��2�2��VB�FW����V�BWf�FV�6R'WB&V���2fV�F�"66�V�Bv�F��WB��FWV�FV�@��W&F��r�WG&�72�&Wf�WvVB6��fW'6Rw27W'&V�B�ff�6���'Ff�Ɩ��B�Ww2vW2v�F��W@�66WF��r&�F�G�R�"&�GV7F���6���2g&����vR���ǒ&W�'F��r�%4�6W"�B4���6V&6�W0�&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�V@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'���B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�@�t#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6P�&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3����VvV��6���V7B�##b���p���VvV��6����V�6���##b��� ��6��fW'6P����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b�����7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�6�VP�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"���FFVB&WWFW'2p�6WFV�&W"�&W�'BF�B5��B��2��B"dB&�GV7F���Ɩ�RB�G2�Wr&V����r��B�@��2F�67W76VB&�7V7F�fR��7F�&vR7W7F��W'2�F�RWfV�B�2&V6�&FVB2&W6V&6��BG&��Т&�GV7F��������B6���W&6���WGWC�&WWFW'26�V�B��BW7F&Ɨ6���W&F��rFFR�"�6��֗F�V�BF�f��V�R��Vf7GW&��r��B5��BF�B��B6���V�B�&R�&Wf�WvVB&WWFW'2r�FW&��ǐ�W��FVB6WFV�&W"r�VvV�&W�'B�BFFVB6�����WfV�Bf�"F�R6���w27FFV�V�G0�V��Vv��6��WF��rWV��V�BF�6F�6g�F��W7F�2FV��B�66V�B�S7�7FV�2�fRV�FW&V@�6���W&6��W6R��66V�B�c7WW&��FR�2���VBf�"WF�B��b&�6W76�'2�BVW&�V��0���FV�FVBWfV�GV�ǒF�6���V7B2��2��R֖�Ɩ���F�W6R&R&W�'FVB6���6���3���VvV�F�B��BF�66��6R66�G���&FW'2�7W7F��W"66WF�6R�WF�Ɨ�F����v�&���B��vW"��&VƖ&�ƗG��"��FW&6���V7B�W&f�&��6RFF�&Wf�WvVB4�TRw27W'&V�B�ff�6��&�GV7BvS�����WrVƖf�VB&�GV7F���֖�W7F��Rv266WFVB�%4�6W"�B4���6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VB��W&`�G&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'��@���fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�@�t#3�ԓ3SU��"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ6�fW&vRF�B��BGf�6R�F�W �6���&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���7��B���B�##b�������VvV��6���V7B�##b���p��6�VP����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b�����7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�F�V@�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W""���FFVB���FWV�FV�@�FV&F�v��b��RV�F�Ɩ�CrE#B�&����GV�RF�BF�&V7Fǒ�FV�F�f�VB�'fV����3SE5 ��B6WfW&����rFWf�6W2�vW"�B���r'G2�F�R&V6�&BW�6�VFW2F�RWF��"w2��fW'&V@��6W"�FV�F�f�6F����BƖ֗G2F�Rf��F��rF�F��2�����W2�6��v�R6��Rv�F���F�66��6V@���Vf7GW&��rFFS��BF�W2��BW7F&Ɨ6�F�R&����b�FW&��2f�"7W'&V�B�r�"�e@�&�GV7F����F�R��FW7B�&�f��R&Wf�Wr&V�V�VBBԆVBw27W'&V�B�ff�6��vR�BF�RW��7F��p���&WWFW'2&W�'B�F�V�&6�f���VBF�R�FW&��ǒV�FW"�&V6�&FVB6���6���2�b��&RF��Sc���V�wRV�G26��VB�FW����V�G2B��&RF��CW�FW&��7W7F��W'27&�72# ���GW7G&�W2�F�R#��66V�W&F�"榗R�#�&6��BW�V7FVBc��B��&�F�FFW2�&�GV7B֗��66WF�6R�WF�Ɨ�F����v�&���BW&f�&��6R��vW"�f'&�6F����B�V��'��&�fV��6R&V���V�F�66��6VB�%4�6W"�B4���6V&6�W2v��&�GV6VB��&WG&�Wf&�R�Wp�&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vVB��W&bG&���p��B��fW&V�6Rcb�&W�6�F�'��VG2�F�R6��6VB��fW&V�6R7V&֗GFW"F�&V7F�'��B��fW&V�6U�w0�7W'&V�BV&Ɩ26�fW&vR����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B7W'&V�Bt#3�ԓ3SU���"ERcr�6�727�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W �6���&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���V�F�Ɩ��Cr�FV&F�v��##b������Ӄ� ��F�V@����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W0�v�F��fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"2�#�FFVB&WWFW'2p�6WFV�&W"#66�V�B�b5��Bw27FFV�V�BF�B�G2f�gF��vV�W&F���E$��Ff�&�V�FW&V@��72&�GV7F���W6��rVG'W�RGFW&��r�B��R��7F�fR�&V��b��F6���BF�@�Gv�#Bv"�DE#U�&�GV7G2�FR��F�R�Ff�&�&R���72&�GV7F����F�R&V6�&BG&VG0�F�W6R26���6���2�B&W6W'fW2&WWFW'2rF�7F��7F���&WGvVV�w&�72F�W2W"vfW"�@�F�R&��'F���76��rf���FW7G3�F�R6���VBB��V7B�SR��&�fV�V�B�fW"F�Rf�W'F�ТvV�W&F����Ff�&�F�W2��BF�66��6R��V�B�&VƖ&�ƗG��7W7F��W"VƖf�6F����"�WGW@�f��V�R�5��B6�B6���W6RWV��V�B��W'26��G&�'WFVBF�7&�F�6�&�6W727FW2'WBF�B��@���RF�V��"W7F&Ɨ6�gV�ǒF��W7F�2&�6W72f��r�F�RWfV�B�26����F�G��E$�֖�W7F��R����BWf�FV�6R�b�$�f��V�R�"W&f�&��6R�%4�6W"�4����B����6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vV@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w27W'&V�BV&Ɩ26�fW&vR��v��6���r��W2'V&�����w6�FRt#3�ԓ3SU��BERcr����F6�VB6��FV��&'�6���W6P�66V�W&F�"&W7V�Bv2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W"6��琢&Wf�WrFFW3���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6R�&V�2vW&R&W6W'fVB�6�W&6U��G3���7��B�sR�G&��##b���# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"B�#��FR�FFVB�Vǒ# �V�v6�V�r�&�&F�'�&W&��BF�B�V7W&W266V�B��"�2��66�V�F�f�2�W&�V�2�Bf�fP�Ɩ6F���7GVF�W2�F�R&V6�&B6GW&W2ff�&&�R�F6�VBV�GV��6��V�F���66S�GV�V@��2���V�V�FF���6���WFVB3�V&�B�3���W"6�&7V�B���B6V6��G2fW'7W2B�26V6��G0�����W6��r7UV�GV���B�6�&W6W'fW2F�RW"w2�W72ff�&&�RWf�FV�6S�fV7F�"��@��V��'��&�V�B�W&�V�2&V���VB6��7G&��VB�6WfW&�6��&�6��2W6VBF�ffW&V�B�v�&�F��2� ����V�V�FF���2��Bv�&���B�7V6�f�2�F�֗�F���v26V�G&��F�R7GVG�W6W2W��'B�Ɩ֗FV@���6��G&��2&F�W"F��7W'&V�Bg&��F�W"�&Gv&R�BF�W2��BFW7B���FV�G&���r� ���fW&V�6R�6��BF�W2��B6��vRF�R�FW7B�fW'7W2��FW7B76W76�V�B�%4�6W"�4����B�����6V&6�W2&�GV6VB��&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBV�6��vVB��W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B��fW&V�6U�w0�7W'&V�BV&Ɩ26�fW&vS����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU���ERcr�"F�R�WvǒƗ7FVB��V��7�7FV�2v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��@�Gf�6R�F�W"6���&Wf�WrFFW2���7F�&�6�&W�'G2�6�'&V7F���2�f�&V67G2�BWf�FV�6P��&V�2vW&R&W6W'fVB�6�W&6U��G3���66V�B�66�V�6R�##b�r�# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#�7FGW3�WFFV@�6���W3���7��@��V�F�Ɩ氢��VvV����WF�����F0�7V��'���Т&�F�RvVV�ǒ�V�W"f�"F�R&Wf��W2gV�ǒ6��6VB���F��7V�F�W&��B�B7&VFVBF�R6���WFP�6WFV�&W"B�#VF�F���v�F��WB&W�6��rF�RV&ƖW"6WFV�&W"rW&��B�F��FFR�"6���WFP�&W�'G2�VF�FVB��6��6V�V7FVB���W&��BWfV�G2�BF�V�"�&�v���6�W&6W2�F�V�&�&�V�FV@�V�vƗ6��B6���W6Rv6V&6�W27&�7266V�W&F�'2��V��'���F�6���FW&6���V7G2�6�gGv&R�@���Vf7GW&��rWV��V�B�F�R7��F�W6�2G&VG2�VvV�w266V�W&FVB##r&�F��&W�'FVB7�7FVТ66�R�BVW&�V�F&vWG226���Wf�FV�6R&F�W"F��FVƗfW&VBv�&���BW&f�&��6R��@���W2�VvV�w2��&�ƗG�F��VWB6���W6RFV��BF�R6�V&W7B66�G�6��7G&��B��BF�7F��wV�6�W0�5��Bw2f�gF��vV�W&F���6����F�G��E$��B�DE#U��72�&�GV7F���6���2g&��FW7FVB��V�B�@��$�&�w&W72��B�G2���VB��B"dBƖ�Rg&��6���W&6���WGWB��BW6W2F�R6��v�RV�F�Ɩ氢CrFV&F�v���ǒ26��R�7V6�f�26����V�BWf�FV�6R�F�R&R�W&��B66V�B66�V�F�f�2Т6��WF��rW"�2&6�w&�V�B&V6W6R�Bv2V&Ɩ2'�F�RW&��BV�C��BF�W2��BV�FW"�Ww5��G0��"W7F&Ɨ6���FV��G&���r�"��fW&V�6R&�G��&V6�V6�VBV�6��vVB��W&bcb�&W�6�F�'��VG0��B7W'&V�B��fW&V�6U�6�fW&vR����F6�VB7W'&V�B6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU�� �ERcrv2fW&�f�VB��B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFR�@�V��FW�VBWf�FV�6R�FV�FW'2���F�f�GV�&V�6��&�'F�f7G2�BV�F�66��6VB&�GV7F����"7W7F��W ��WG&�72�6�W&6U��G3���66V�B�66�V�6R�##b�r�# ��6��R��F�72�##b�����7��B�sR�G&��##b���# ��7��B���B�##b������V�F�Ɩ��Cr�FV&F�v��##b�������VvV��6���V7B�##b���p����fW&V�6W����WF��G&��##b���@����W&b֖�fW&V�6P����W&b�G&���p��FFS�##b���# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т�������F�V@��ƖvV�@�7V��'���Т67&VV�VB��&��&�G��&�f��W2�BF�R�W�BF�&VR��FW7B�&Wf�WvVB����&��&�G�&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r6WFV�&W"R�#"�FFVB&WWFW'2r6WFV�&W"# �&W�'BF�BƖ&&V�fV��VBBԆVBw2��V�wRc��B��rF&vWG2##r�72&�GV7F����@�6���W&6��&V�V6R�V&ƖW"F��F�R2##r&�F�&W�'FVB�����&W6W'fVBF�B��FW"FFV@�&�F��B&V6�&FVBF�R�WrF�&VR�F��W2�Ӄ�W&f�&��6R�BS��6��6�W7FW"6���20�&W�'FVBF&vWG2&F�W"F��v�&���BWf�FV�6S�&V6�6������FV���vW"�F����w��&V6�fW'���f'&�6F����B�V��'�FFvW&R��BF�66��6VB�FFVBƖvV�BFV6�����v�W22G&6�VB��F��70�7WƖW"gFW"�G2��BR�cr&��Ɩ�����r���r���v�F����VB���6F���2�b&�WBS"�R�b�W@�&�6VVG2F�f7FW"�F�6��&�GV7B�B6��"dB�B#R�RF�66�G��F�R&V6�&B6W&FW26�F��&�6VB�BfV�F�"�Ɨ7FVBCr�r��eB�'Ff�Ɩ�g&��fW&�f�VB��v��7VVB6���V�B֗����V�B��7W7F��W"VƖf�6F����BF��W7F�26����V�B&�fV��6R�%4�6W"�4����B����6V&6�W2&�GV6V@���&WG&�Wf&�R�Wr&��'�Wf�FV�6R�6�F�V�"&Wf�WrFFW2F�B��BGf�6R�&V6�V6�VBF�RV�6��vV@���W&bG&���r�B��fW&V�6Rcb�&W�6�F�'��VG2�B7V&֗GFW"F�&V7F�&�W2�W2��fW&V�6U�w0�7W'&V�BF6�&�&C����F6�VB6��FV��&'�6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr� ��V����\;�v2fW&�f�VB�6V&6����ǒ&��&�G�6�fW&vRF�B��BGf�6R�F�W"6���&Wf�WrFFW2�6�W&6U��G3���Ɩ&&�c��##b���# ��ƖvV�@��ƖvV�B֗��##b���# ����W&b�G&���p����W&b֖�fW&V�6P����fW&V�6W���FFS�##b���#0�7FGW3�'F���6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��6��Т������7V��'���Т67&VV�VBF�W6RR6���W2v�F�6WFV�&W"b�#2�fW&����BgV��6�W&6R&WfƖFF����`�WfW'�&�f��R�&6�f���VB�VvV�w2&��'�FW����V�B�FW7F��rF�66��7W&R�B��FW&6���V7BMx㝭�G����ƭy��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��&�&V��wV�v�p�����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2��FFVB&�&V�w2�7F�&W"��&��'�f�Ɩ�rF�B�G2V���VB�W�B�vV�W&F���&�GV7B6���WFVB'&��r�W�BV�FW&V@��7B�6�Ɩ6��fƖFF����F�R6���W�V7G27W7F��W"6�Ɩ�r�Bf��V�R&�GV7F����fW"F�P�6�֖�rV'FW'2�7V&�V7BF�6F�6f7F�'�fƖFF����BVƖf�6F���F��2�2F�W&Vf�&R�V�v��VW&��r֖�W7F��R&F�W"F��7W7F��W"F�F����VƖf�6F����"&�GV7F�����6�&V6�&FVB&�&V�w2&W7B�Vff�'G2�6V�V�B�b3֖�Ɩ���6�&W2B��C3��V6��f�"W7F��FV@��WB&�6VVG2�b��CB��r&��Ɩ���F�Rf�Ɩ�r���6FW2sRF�7Wǒ�6���&�7W&V�V�B��&�GV7F���&VF��W72�B6���W&6�Ɨ�F����#RF�&W6V&6��6�gGv&R�BV6�7�7FV�v�&���@�RF�v�&���r6�F��'WBƗ7F��r&�f�&V���VB&W&WV�6�FR�BF�RG&�67F���֖v�@���B&�6VVB�2�b6WFV�&W"3F�R6����BW6VB��C��"�֖�Ɩ����"BR��b�G2&��"�Vǐ��6V�V�B&�6VVG2�F�Rf�Ɩ�rF�B��B��RF�R�Wr&�GV7B�"F�66��6R7V6�f�6F���2��f'&�6F�����V��'����V�B�7W7F��W'2�&V�6��&�2�7WƖW'2�&W6W'fVB66�G��"6���W&6���6��֗F�V�G2���ǒ&�&V�w2&�f��RFFRGf�6VB�7W'&V�BV&Ɩ2&V6�&G2f�"%4�6W"�wV�v�p��B���w6V֒F�B��BW7F&Ɨ6���F�W"VƖf�VB&�GV7F����FW����V�B�7W7F��W"�F�F���� �&V�6��&�֖�W7F��S�6V&6����ǒ6�fW&vRF�B��BGf�6RF�V�"&Wf�WrFFW2���W&bG&���p�cb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���V@�BF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���Wr�F6�VB6���W6P�66V�W&F�"6��&�6��v��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���&�&V���6V�V�B�##b�����&�&V��wV�v�p�����w6V����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b���7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p����WfF �����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"B��FFVBF�R�7F�&W"��F������Wr�VƗG��&�GV7F�fR�f�&6W2wV�F�6R26�����WWG&��Ɩ7�WfV�B��B6��2f�"f7FW ��F�V�'��B6�&R�FV6�����w�FWfV���V�B�7G&��vW"6��WFR��v�&�F���BFF7Wǒ���F������FVw&FVB6��WF��r��vW"�WGv�&��B'&�FW"�F�F���7&�72��GW7G&�W2�@�FWf�6W2�F�R6�RwV�F�6R&WV�&W2��6�ǒ&�&�FR����r�Bv&�2v��7B�W&B�Ɩ�P�6��WF�F����'V&&�W2�&Ɩ�B��fW7F�V�B�BGWƖ6F�fR��6�W��6�����B6WG2�Ɩ7��F�&V7F���'WB&�f�FW2��V�F�f�VB'VFvWB����V�V�FF���66�VGV�R�&�7W&V�V�Bv&B� �6������6F������g&W6�VƖg���r&�GV7F����FW����V�B�7W7F��W"�F�F����f���6��p��"&V�6��&�֖�W7F��Rv2W7F&Ɨ6�VBf�"F�R67&VV�VB6���W2�6���&�f��R&Wf�WrFFP�Gf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���F6�VB6���W6R66V�W&F� �6��&�6��v��7B7W'&V�B'V&���t#3�ԓ3SU��ERcr�"�V����\;�7�7FV�2���F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���6�����Wr�VƗG��&�GV7F�fR�f�&6W2��Ɩ7��##b�������W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b�� �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p����WfF �����w6V�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"R�"�&6�f���VB��WfF"w0�6WFV�&W"#��ff�6����FW&��&W�'C�f�'7Bֆ�b&WfV�VRv2$�"�CR�r֖�Ɩ���uuR&WfV�VP�v2$�"�b�֖�Ɩ���BF�R6���&W�'FVB��&RF��3�7W7F��W'2���fW&V�6R�&�GV7@�&WfV�VRv2$�"cSB�"֖�Ɩ���BG&���r�&�GV7B&WfV�VRv2$�"#c�֖�Ɩ���F�Rf�Ɩ�p�F�W2��B��R7W7F��W'2�"F�66��6RV�B6���V�G2�&�GV7B֗��WF�Ɨ�F����"v�&���@�&W7V�G2�F�F�w&�72�&v��fV��F�r�"R��WB&�f�Bv2$�"b�2֖�Ɩ�����w6�FR$� �sc�"֖�Ɩ��f�"�f�VRv����W&F��r66��WFf��rv2$�""�s�b&��Ɩ���B&�'&�v��w0�&V6�VB$�""�c�r&��Ɩ�����6��WF��r�6��WF���2&WfV�VRfV��c��R֖B7Wǒ�&W6�W&6P�6��&F��F����BFVƗfW'��66�VGV�RF�W7F�V�G3�F��2F�W2��B'��G6V�bW7F&Ɨ6�7V6�f�0�6��6��'FvR�"7W7F��W"6�6V��F������ǒ��WfF"w2&�f��RFFRGf�6VB�7W'&V�BV&Ɩ2&V6�&G2f�"F�P��F�W"67&VV�VB6���W2F�B��BW7F&Ɨ6���F�W"g&W6�VƖg���r&�GV7F����FW����V�B��7W7F��W"�F�F����f���6��r�"&V�6��&�֖�W7F��R���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��W&b��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7WƖVB���F6�VB6���W6R66V�W&F� �6��&�6��v��7B7W'&V�B'V&���t#3�ԓ3SU��ERcr�"�V����\;�7�7FV�2���F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����WfF"�##bփ�&W�'@��wV�v�p�����w6V����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b�� �7FGW3�WFFV@�6���W3���&�&V���VvV��7V��'���Т&�F�RvVV�ǒ'&�Vf��r�V�W"f�"F�RgV�ǒ6��6VB�7F�&W"R�W&��B�B&WF��VB���V&ƖW"VF�F���2�VF�FVB��f�W"6���6����W&��BWfV�G2�BF�V�"�&�v���6�W&6W2��W6��rF�R�'6W'fF�'�w2F�ǒ�fW&���rV�vƗ6��B6���W6R6V&6�W27&�72b&��&�G��@���FW7B�&Wf�WvVB6���W2F�6�V6�v2�F�R6���WFRVF�F���6W&FW2&�&V�w2&W�'FV@��7B�6�Ɩ6��'&��r�Wg&��fƖFF����7W7F��W"6�Ɩ�r�VƖf�6F����Bf��V�R&�GV7F��㰢�B�6�6W&FW2&�&V�w2&��6VB��CB��r&��Ɩ���6V�V�Bg&��6���WFVBgV�F��r��&W6W'fVB66�G��B&VƗ�VB�WGWB�F�R�F����wV�F�6R6��&��W27G&��vW"�6��WFR���v�&�F���FF�BF�F���7W�'Bv�F�6��G&��2v��7B'V&&�W2�&Ɩ�B��fW7F�V�B�@�GWƖ6F�fR��6�W��6����'WB&�f�FW2��'VFvWB�66�VGV�R�&�7W&V�V�Bv&B�"6��琢���6F�����VvV��BV�6���w27&�72�Ɩ6V�6R�2&V6�&FVB2v��&��W�6��vR&F�W"F��6���7Wǒw&VV�V�B�"&��b�b��VB66V�BFWV�FV�7����W&bG&���rcb��@���fW&V�6Rcb��VG2&V���VBV�6��vVB��B��fW&V�6U�7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒVF�FVB���FF�F�����FW&�����W&��BWfV�Bv0�W7F&Ɨ6�VB�B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFP�FW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R����V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���&�&V���6V�V�B�##b�����6�����Wr�VƗG��&�GV7F�fR�f�&6W2��Ɩ7��##b������VvV��V�6����FV�B�Ɩ6V�6R�##b��P���VvV��V�6����FV�B�Ɩ6V�6R�&WWFW'2�##b��P����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��0�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p�����w6V���V��V熖��F�V@�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"b�2�BԆVBv2�6�WFFV@�&V6W6RF�R66WFVB�&�WB6�W&6RF�&V7Fǒ6�fW&VB�B�&WWFW'2&W�'FVB�D2f�'7Bֆ�bFF�6��v��r�VvV�6��VBSs"��66V�W&F�'2��6���f�"#�R6�&R�v�F�BԆVBBrR��6�'&�6��BRR�B�V��V熖�BBR�F�F�6���V�G2vW&R"��֖�Ɩ����B����uRV�G2&�6P��V&ǒ�RF��b֖�Ɩ���F�Rf�wW&W27G&V�wF�V�F�F���Wf�FV�6R'WBF���B�FV�F�g��&�GV7B֗��'W�W'2�66WFVBFVƗfW&�W2�WF�Ɨ�F����v�&���G2�&�6W2�"6����V�@�&�fV��6R��BW��'B&W7G&�7F���2��R�&�WB6�&RV�7V�F&�R2g&��F�W"�W&f�&��6P�6��&�6����V��V熖�w2�7F�&W"&��'�&V�V6R�B$Ɩ�bw2&V�V6R�c�B&W�6�F�'�FFV@���7V7F&�R�&6�V�BWf�FV�6R7&�726��v�R��B�V�F����FR��FW2�&�66�VGVƖ�r��44��6���V7F�fW2�B��RĔ$U$�&W&�GV7F����fV�F�"F6��B7��6�&�旦F���&W7V�G2�6���F6�VBf�&V�v�ֆ&Gv&R6��G&���F�66��6VB�vW"�"6�7B���FWV�FV�B&W&�GV7F���� �&�GV7F����7W7F��W"Wf�FV�6R����w6V֒��vV�V�B&W�'FVB�FW&���&FW"w&�wF��&�GV7F���&W77W&R�66�G�W��6����B6��gBF�v&Bg&��B�V�BG&6��B6�V��rWV��V�C����&FW"��WGWB�66WFVB�F����f"���FR�"VƖf�6F���6�V�G2vW&RF�66��6VB��BgV�FV@�W��6���&��V7B&V���2FV��VB���ǒ�VvV��6�'&�6����V��V熖��BԆVB�B���w6V�&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&B7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���6������66V�W&F�"�6�&R�&WWFW'2�##b��0���V��V熖��&Ɩ�b�##b�� ��&Ɩ�bַV��V熖��7W�'@�����w6V֒�66�G��##b�� ��GWb�g@��wV�v�p����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��@�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ���WF������&R�F�&VG0��6��fW'6P�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"r�B�&6�f���VB5��Bw26WFV�&W �#�&�&B�&�fVB&��6�F�W6R$�"2�&��Ɩ���b�fW"�&�6VB��gV�G2F�v&B$�"#B��&��Ɩ��FV6�����w��&W6V&6�&��V7B�B$�"R�&��Ɩ��F�v&B$�"�&��Ɩ��6V6��B��6P�E$�vfW"&6��V�BFW7F��r&6R�F�R&��6�&V���VB7V&�V7BF�6�&V���FW"&�f��@�F�W2��BW7F&Ɨ6�6��7G'V7F������7F��VBFW7B66�G��VƖf�VB�WGWB���V�B�"�$Тf��V�R��T2w2�7F�&W"2f�Ɩ�rFFVB&��6VB$�"3R֖�Ɩ����fW7F�V�B����v�R���w�V�v�F���$�"#S֖�Ɩ��&�V�B��FV�FVB'Fǒf�"��'&�B�&��F��r�BF�֖2�&V�WV��V�@�&W6V&6��B7W7F��W"FVƗfW'��F�Rw&VV�V�B&V���VBV�6�v�VB��B&W�'FVB7W7F��W �fƖFF���F�W2��BW7F&Ɨ6�66WFVB&�GV7F���F���2�F�&�Vv�WB�&VƖ&�ƗG��"F��W7F�0�6����V�B&�fV��6R���ǒ5��B�B�T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2����F�W"g&W6��VƖg���r&�GV7F����FW����V�B�7W7F��W"�F�F����"f���6��r֖�W7F��Rv2W7F&Ɨ6�VB���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6P�cb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&@�7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B� ���7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�V@�6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F�����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���7��B��fW&gV�B�&��V7G2�##b���#����V2���v�R֦��w�V�֖�fW7F�V�B�##b��0����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��P�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p���WF������&R�F�&VG0�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"��R�&6�f���VB6�V�FW'���Bw0���FWV�FV�B6V6��B�V'FW"��B6���V�BW7F��FS���D2�V�BBR'�f��V�R�&涖�rF��&@�&V���B6�7V�r�B4���旂�v�F�6���V�G2W#"R�V"�fW"�V"�BRRV'FW"�fW"V'FW"�6�V�FW'���B&�VB��D2f�gF�'�&WfV�VR&V6W6R�G2֗�&V���VB6��6V�G&FVB����vW"�&�6V@�6��7V�W"Ɩ6F���2v�F���r6�&R�bFF�6V�FW"V�FW'&�6R54G2�6�F�Rf��V�R&涖�p�F�W2��BW7F&Ɨ6�V�FW'&�6R�VFW'6���&�GV7F�����V�B�7W7F��W"VƖf�6F����"7WǒТ6���6�fW&V�v�G���6�FFVBF�R6��v��7F�6�W�6��vRw2VwW7B#&V6�&B66WF��r��D2w0�Ɩ6F���f�"&��6VB$�"32&��Ɩ��5D"�&�WB&�6R�F�R&��V7Bv2V�FW"��V�'���F�R&V6�&BF�W2��BW7F&Ɨ6�6���WFVBf���6��r�f���&�6VVG2���7F��VB66�G���VƖf�VB�WGWB�"gV�ǒF��W7F�2WV��V�B�B�FW&��2f��r���ǒ��D2&V6V�fVB�Wp�&�f��R�&Wf�WrFFR�7W'&V�B�ff�6���"�&�v����6�W&6R6�V6�2f�"F�R�F�W"67&VV�V@�6���W2F�B��BW7F&Ɨ6���F�W"g&W6�VƖg���r&�GV7F����FW����V�B�7W7F��W"ТF�F����"f���6��r֖�W7F��R���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&B7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����F2�7F"��&�WB֗��##b���#��6�V�FW'���Bז�F2�"�6���V�G2�##b��� ����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��`�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p���WF������&R�F�&VG0�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"��b��WF�w2�7F�&W"bf�Ɩ�p�f�&V67B��V'��6WFV�&W"&WfV�VR�b$�""�SC�2�b&��Ɩ���BGG&�'WF&�R�WB&�f�B�`�$�"S�b��sr�"֖�Ɩ���6��&VBv�F�$�"3CR�R֖�Ɩ����72�V"V&ƖW"�F�Rf�&V67@�v2V�VF�FVB��BF�Rf�Ɩ�r6�B��7&V6VB��fW7F�V�B��7Wǒ�6���6���W2�6�6��G&�'WFVBF�F�R&�f�BGW&�&�V�C�7W7F��W"��W2�V�G2�v�&���G2�WF�Ɨ�F�����f'&�6F����B�V��'�&�fV��6RvW&R��BF�66��6VB��U$w27V'6�F�'�6�v�VB�w&VV�V�@�F�7V�&R��'&�B�&��F��rWV��V�BFWfV��W"�V��R���vV�f�"$�"CB�b֖�Ɩ���6��6��p�6��F�F���2&V���VB��B&W�'FVB&�GV7BfƖFF����6���V�G2�B&WVB�&FW'2F���@�W7F&Ɨ6���VB�7W7F��W"66WF�6R�&�GV7F���W6R�F�&�Vv�WB���V�B�WF��R�"F��W7F�0�6����V�B&�fV��6R�6�V�FW'���B&W�'FVBF�B��D2v26��GV7F��r"dB�BFW7F��r�b�6V6��B�vV�W&F���bv"�DE#RFWf�6RB&�Vv�ǒR�vfW'2W"���F�v�F����72�&�GV7F�������G2'F�6�R�6�FW67&�&VBb�RR7W'&V�B��B6���V�B6�&Rv�F��WBV&Ɩ6ǐ�&V6��6�Ɩ�rF�B&6�2v�F�F�RV&ƖW"BR6V6��B�V'FW"W7F��FR�6�&�F�vW&R&WF��V@�v�F�F�RF�67&W�7�F�66��6VB���ǒ�WF���U$�B��D2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6P�cb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w27W'&V�BF6�&�&@�7F���7WƖVB���F6�VB6���W6R&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B� ���7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�V@�6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F�����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3����WF����R����F��f�&V67B�##b��`���W&��V��R֦��vV��7V�6�F����##b��`��6�V�FW'���Bז�F2֗��G&��##b��0����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��p�7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p�����&R�F�&VG0��6��fW'6P�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"�r�5��Bw2�7F�&W"rf�Ɩ�p�f�&V67B��V'��6WFV�&W"&WfV�VR�b$�"#cR��#s�&��Ɩ���BGG&�'WF&�R�WB&�f�B�`�$�"3��3"�R&��Ɩ���6��&VBv�F�$�"3"��B&��Ɩ���b&WfV�VR�B$�"R�#�&��Ɩ����72�V"V&ƖW"�F�Rf�wW&W2vW&R&VƖ֖�'��BV�VF�FVB�5��BGG&�'WFVBF�R6��vP�F���v�W"E$�&�6W2���v�WF�Ɨ�F������v�W"�WGWBW"vfW"�66�R�B&�GV7B�֗����&�fV�V�G2�'WBF�B��BF�66��6RV�G2�&�GV7B��WfV�&WfV�VR���V�B�7W7F��W"֗���$Тf��V�R�"F��W7F�2WV��V�B�B�FW&��2&�fV��6R�&6�f���VBF�RwW���V�6��6V�FW"w0�&W�'FVB�W&F��r֖�W7F��S���"��EB3S6&G2���6R��RvW&R�W&F��r�vw&VvFP�FV��Bv2�V"gV����B�B��&RF��#7W7F��W'2vW&RFW7F��r�"W6��rF�R6�W7FW"�F��0��2��VBFW����V�B�BF�F���Wf�FV�6R���Bv�&���B��WfV�WF�Ɨ�F����&VƖ&�ƗG����vW"�"�F6�VBf�&V�v�ֆ&Gv&R&V�6��&��F�R6���VBgV���7F6�F��W7F�27FGW2�6��6�2V&Ɩ26����V�B��WfV�&�fV��6R���ǒ5��B�B���&RF�&VG2&V6V�fVB�Wr&�f��RТ&Wf�WrFFW2���W&bG&���rcb�&V���VBBV&c#6v#&3c#��fc�sF36CfffcCC#��B��fW&V�6Rcb�&V���VBBF&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�w0�7W'&V�BF6�&�&BƗ7FVB'V&���t#3�ԓ3SU��ERcr�B�V����\;�7�7FV�2'WB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�B���F�f�GV�&V�6��&�'F�f7G2vW&R��@�W��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��r6��vVB�6�fW&vR&V���0���6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W �66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3���7��B���R����F��f�&V67B�##b��p�����&R�F�&VG2�wW���6V�FW"�##b��� �����&R�F�&VG2�wW���6V�FW"�&��'��##b��`����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�����FW&���6��vP�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"���F�R7vVWf�V�B���Wvǐ�fW&�f�VB6�W&6RW7F&Ɨ6���r�FW&��6��vR��VƖf�VBFVƗfW&�W2�&�GV7F���f��V�R� ���V�B���VB�7W7F��W"66WF�6R��$�&�fV��6R�F���FW����V�B��"7W'&V�B66V�W&F� �W&f�&��6R�6V6��F'�6�fW&vR�b�T2w2F�&VR�F��f�fRזV"WV��V�B�FWfV���V�B��F�@���BW7F&Ɨ6�6���WFVBF���2�7W7F��W"VƖf�6F����"&�GV7F���W6R&W���BF�RW��7F��p�&V6�&B�&��6VB6���W6RFV6�����w��W��'B6��G&��2vW&R��BGG&�'WFVBF�G&6�VB6���w0��W&F��r6��vR���6���&�f��R�&Wf�WrFFRGf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�����FW&���6��vP�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W""���F�R7vVWf�V�B���Wvǐ�fW&�f�VB6�W&6RW7F&Ɨ6���r�FW&��6��vR��VƖf�VBFVƗfW&�W2�&�GV7F���f��V�R� ���V�B���VB�7W7F��W"66WF�6R��$�&�fV��6R�F���FW����V�B��"7W'&V�B66V�W&F� �W&f�&��6R�5��Bw26��&��&��6VB$�"��"�&��Ɩ��6�&R'W�&6�f�"V����VP���6V�F�fW2�"6�F�&VGV7F����'WBF�R&��6�&V���VB7V&�V7BF�&�&B7F����BF�@���Bf���6R�"W7F&Ɨ6�&�GV7F���66�G��VƖf�VB�WGWB�"�$�f��V�R�&��6V@�6���W6RFV6�����w��W��'B6��G&��2vW&R��BGG&�'WFVBF�G&6�VB6���w2�W&F��p�6��vR���6���&�f��R�&Wf�WrFFRGf�6VB���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����\;����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3�����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b����7FGW3�WFFV@�6���W3����V0��6�'&�6���7��@���VvV����V��V熖���WF����W&��F�V@����F0�7V��'���Т&�F�RvVV�ǒ'&�Vf��r�V�W"f�"F�RgV�ǒ6��6VB�7F�&W""��W&��B�B&WF��VB���V&ƖW"VF�F���2�VF�FVB��6��6���6�F�66��7W&W2V&Ɨ6�VB��F�RW&��B�BF�V� ��&�v���6�W&6W2�W6��rF�R�'6W'fF�'�w2F�ǒ�fW&���rV�vƗ6��B6���W6R6V&6�W0�7&�72"&��&�G���Bf�W"��FW7B�&Wf�WvVB&�f��W2�W2f���7&�72�6V7F�"v6V&6��&WWFW'2�&W�'FVB�D26���V�BFF7G&V�wF�V�VBWf�FV�6R�bF��W7F�266V�W&F�"F�F���'W@�F�B��BW7F&Ɨ6�&�GV7B֗��66WFVBFVƗfW&�W2�WF�Ɨ�F����"�F6�VBW&f�&��6R�5��@��B�WF�7WƖVB&VƖ֖�'�V�VF�FVBWf�FV�6R�b6���W&6��66�Rv�F��WBF�RV�B��7W7F��W"���V�B�v�&���B��$��f��V�R�"&�fV��6RFWF���VVFVBf�"&�GV7F���6��6�W6���2��U$w26�v�VB7V�6�F����B�T2w2&��6VB��fW7F�V�BF&vWFVB��'&�B�&��F��rWV��V�B��'WB6��6��r�&�GV7F���W6R�B�W&F��rWf�FV�6R&V���VB��6���WFR�6�V�FW'���BFW67&�&V@���D2w2E$�v�&�2&W6V&6��BFW7F��rv�F��WB�72�&�GV7F�����f�"F�RFW7FVBFWf�6R���W&bG&���rcb��B��fW&V�6Rcb��VG2&V���VBV�6��vVB��B��fW&V�6U�7WƖVB���F6�VB7W'&V�B6���W6R66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V�����\;����F�f�GV�&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒVF�FVB���FF�F�����FW&������W&��BWfV�Bv2W7F&Ɨ6�VB�B��FFVBf�&V67B6��vVB�6�fW&vR&V���2��6���WFRf� �&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2�FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�B6����V�B&�fV��6R�6�W&6U��G3����V2���v�R֦��w�V�֖�fW7F�V�B�##b��0��6������66V�W&F�"�6�&R�&WWFW'2�##b��0��6�V�FW'���Bז�F2֗��G&��##b��0��7��B���R����F��f�&V67B�##b��p���WF����R����F��f�&V67B�##b��`���W&��V��R֦��vV��7V�6�F����##b��`����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W���FFS�##b��# �7FGW3�WFFV@�6���W3����6�V�v����V0��6�'&�6���7��@��V�f��P��V�F�Ɩ氢��VvV�������Ɩv�@���W&��6֖0����F0���VƖ�w6�V�p��'6�6W ��wV�v�p��6��fW'6P��6�VP�7V��'���Т67&VV�VB��"&��&�G��&�f��W2�BF�Rf�W"��FW7B�&Wf�WvVB&V����r&�f��W2v�F���fW&���rV�vƗ6��B6���W6R6V&6�W26�fW&��r�7F�&W"2�#�F�V�&�F&vWFVB&6�f����6V&6�W27&�72�V��'��66V�W&F�'2�BWV��V�B�FFVB5��Bw2�7F�&W"���V�6V�V�B�b�6V�b�FWfV��VBDc"fW'F�6��7W'&�V�F��r�vFR6V��&6��FV7GW&R�B���VB6W'fW"DE#R$D��Т'��V"�V�B##b�F��2�2&�F����B6���WFVB&�GV7B�VƖf�6F����"f��V�R�&�GV7F���֖�W7F��S�7VVB�66�G����FR�7W7F��W'2�B��FV�FVBf��V�RvW&RV�F�66��6VB�&6�f���V@��7F�&W""W���&�F���&W�'F��r��V�f��Rw2U4�cB�7WW&��FRv�F��'F��v���V�V7G&�6��@��F�6�66�R��WB��FW&6���V7G2�v�F��WBG&VF��rF�7��2��W&F��rFW����V�B� �W&f�&��6R&W7V�B�F�R6�R&W�'BFFVB�T2w26���VB66�R�b��&RF���Ã7V�V�F�fP�&V7F���6��&W'2���fW"##&�GV7F���Ɩ�W2�B�G2�Wvǒ�V�6�VB��v��7V7B�&F��WF6���B3����ǖ&FV�V��B7�7FV�2�7W7F��W'2�&�GV7B֗����FW2�WF�Ɨ�F����WF��R���V�@�6��G&�'WF����BVƖf�6F����bF�R�WrF���2&V���V�F�66��6VB���ǒ5��B�V�f��R�@��T2&V6V�fVB�Wr&�f��R�&Wf�WrFFW2���W&bG&���rcb�&V���VB@�V&c#6v#&3c#��fc�sF36CfffcCC#�B��fW&V�6Rcb�&V���VB@�F&#c66C#�V#3f&S##&3��S3�3cf3�sSS6b���fW&V�6U�7F���7WƖVB���F6�VB7W'&V�B6���W6P�66V�W&F�"&W7V�Bv��7B'V&���t#3�ԓ3SU��ERcr�"�V����V�����F�f�GV��&V�6��&�'F�f7G2vW&R��BW��W7F�fVǒ&Wf�WvVB���FFVBf�&V67B�"��7F�&�6�'&�Vf��p�6��vVB�6�fW&vR&V���2��6���WFRf�"&�fFRFW����V�G2�V��FW�VB6���W6RF�66��7W&W2��FV�FW'2�7W7F��W"66WF�6R�&�GV7F���f��V�R���V�B�WF�Ɨ�F����&VƖ&�ƗG���vW"�@�6����V�B&�fV��6R�6�W&6U��G3���7��B�Fc"�&F����##b����wV�����W���##b�� ����W&b�G&���p����W&b֖�fW&V�6R�b�����fW&V�6W�
- Added: CXMT 4F2 architecture records technical breakthrough; next-generation DDR5 RDIMM expected by year-end
- Added: On the ground at the 2026 Guoxin Expo as semiconductor companies display new products
- Updated: Continuous inference benchmark dashboard
1 field changes
- reviewed_at: 2026-10-17 → 2026-10-20
- Updated: MLPerf Inference v6.1 closed submissions
1 field changes
- reviewed_at: 2026-10-17 → 2026-10-20
- Updated: MLPerf Training v6.0 submissions
1 field changes
- reviewed_at: 2026-10-17 → 2026-10-20
2026-10-19b08e026 ↗
briefing: publish October 12-18 weekly report
- Added: Adoption and earnings scale while sovereignty proof stays incomplete
- Updated: data/research-state.yaml
3 field changes
- last_run_status: no-material-change → updated
- coverage: Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. No newly verified source established a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's multi-year equipment-development plan did not add completed products or production qualification beyond the existing record. CXMT's chairman-proposed share buyback remained subject to board action and did not finance or establish production capacity. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 and Inference v6.1 heads were unchanged, while InferenceX still supplied no matched current Chinese accelerator result. No dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance. → Published the complete October 12-18 weekly briefing after auditing six canonical disclosures and their original sources, using the daily overlapping English and Chinese reviews of 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. The report distinguishes IDC shipment share from performance evidence, preliminary CXMT and MetaX financial forecasts from audited operating detail, hybrid-bonding transactions from production qualification, and YMTC's DRAM research wafers from commercial output. No additional material in-period event was established. MLPerf Training v6.0 and Inference v6.1 heads were unchanged, while InferenceX still supplied no matched current Chinese accelerator result. No dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"updated","companies":["amec","cambricon","cxmt","huawei","kunlunxin","metax","naura","thead","ymtc"],"summary":"Ran the weekly briefing helper for the fully closed October 12-18 period and retained all earlier editions. Audited all six canonical disclosures published in the period and their original sources, using the Observatory's daily overlapping English and Chinese searches across 12 priority-1 and four oldest-reviewed profiles plus a final cross-sector gap search. Reuters-reported IDC shipment data strengthened evidence of domestic accelerator adoption but did not establish product mix, accepted deliveries, utilization or matched performance. CXMT and MetaX supplied preliminary unaudited evidence of commercial scale without the unit, customer, yield, workload, HBM-volume or provenance detail needed for production conclusions. NAURA's signed acquisition and AMEC's proposed investment targeted hybrid-bonding equipment, but closing, production use and operating evidence remained incomplete. Counterpoint described YMTC's DRAM work as research and testing without a mass-production plan for the tested device. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["amec-qinghe-jingyuan-investment-2026-10-13","china-ai-accelerator-share-reuters-2026-10-13","counterpoint-ymtc-ipo-dram-2026-10-13","cxmt-nine-month-forecast-2026-10-17","metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]}]
2026-10-19e7e934a ↗
research: record October 19 no-change sweep
- Updated: data/benchmarks.yaml
2 field changes
- assessment: Follow-up through October 18 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads and submitter listings plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability. → Follow-up through October 19 did not verify a matched independent comparison of current Chinese accelerators against Rubin, GB300, MI355X, TPU v7, or OpenAI Jalapeño. The review checked MLPerf Training v6.0 and Inference v6.1 repository heads and submitter listings plus InferenceX's published coverage, not every result artifact. InferenceX independently analyzed a 16-rank Ascend 950DT DeepSeek V4 profiler trace and confirmed functioning optimized day-one inference, but did not publish its planned same-harness H200/B200 numerical comparison. MLPerf's standardized results are submitter-run and are not automatically independent measurements. MLPerf Inference v6.1 advanced to commit 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still contain no current Chinese accelerator vendor. A Pengcheng Laboratory preprint adds measured Ascend 910C scientific-computing results against A800/H800 controls, but it is not a model-training or inference comparison and does not use current frontier foreign hardware. Absence of a verified public result is not evidence of incapability.
- reviewed_at: 2026-10-18 → 2026-10-19
- Updated: data/research-state.yaml
4 field changes
- last_research_at: 2026-10-18 → 2026-10-19
- last_automated_run_at: 2026-10-18 → 2026-10-19
- coverage: Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. No newly verified source established a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's multi-year equipment-development plan did not add completed products or production qualification beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 and Inference v6.1 heads were unchanged, while InferenceX still supplied no matched current Chinese accelerator result. No dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance. → Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. No newly verified source established a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's multi-year equipment-development plan did not add completed products or production qualification beyond the existing record. CXMT's chairman-proposed share buyback remained subject to board action and did not finance or establish production capacity. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 and Inference v6.1 heads were unchanged, while InferenceX still supplied no matched current Chinese accelerator result. No dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, individual benchmark artifacts, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.
- runs: [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]}] → [{"date":"2026-09-12","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","biren","ciomp"],"summary":"These 15 companies were screened, not all fully revalidated. Read original CIOE reporting and optical suppliers' public sites; added reported shipment progression for InnoLight and Eoptolink. Read Biren's dated primary announcements; added software-validation and research-laboratory events, preserving the distinction between publication and event dates. Reviewed the indexed original September 11 Reuters Enflame report and added planned use of IPO proceeds without duplicating its existing IPO event. Only these four company review dates advanced. Priority-company searches also revisited the September 10 HBM pricing report and July DUV reporting; no additional chip-performance or qualified-tool claim was accepted. AMEC's official site and the full Financial Times DUV article could not be retrieved. Follow up AMEC's six-tool announcement, the DUV testing chronology, reported optical-module export restrictions, and memory/equipment filings; do not infer qualification from headlines. Downloaded and visually checked Biren's primary half-year infographic; added its revenue, loss, R&D and unnamed-customer delivery claims with vendor attribution, not an independent audit. ACM and CIOMP searches did not establish a verified new qualifying milestone; their profile dates remain unchanged. Checked MLPerf Training v6.0 repository listings, Inference v6.0 closed-submitter listings via GitHub, and InferenceX's public coverage. No matched current China-versus-frontier result was verified; individual result artifacts were not exhaustively audited. Historical news, period briefings, forecasts and accelerator specifications are retained. Research scope is partial independently of GitHub sync or Cloudflare publication status.","source_ids":["cioe-optics-2026-09-11","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","enflame-ipo","hbm-reuters","duv-reuters","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","acm","ciomp","empyrean"],"summary":"Screened all priority-1 profiles and the three oldest-reviewed non-priority profiles with overlapping English and Chinese searches. Added PyTorch Foundation primary evidence that Cambricon joined as a Platinum member with Governing Board and Technical Advisory Council representation and reported upstream work across compile, runtime and distributed areas. Added Huawei's existing Platinum membership and co-chair role in the Foundation's Accelerator Integration Working Group. These are ecosystem and governance signals, not performance results. Closed the prior AMEC follow-up using detailed Science and Technology Daily reporting on six CSEAC launches across etch, PECVD, ALD, metal-film deposition and silicon-carbide epitaxy; kept specifications distinct from customer qualification and volume-production evidence. Reviewed current official portfolio and news pages for ACM Research Shanghai, CIOMP and Empyrean; no new qualifying production, customer or complete-flow milestone was verified, so their facts were unchanged while their actual review dates advanced. Rechecked the Financial Times DUV story, HBM pricing coverage, the September 11 optics report and Enflame's IPO follow-up during the overlap window; no additional claim was accepted. Reviewed the FCC's final component-rule coverage but did not attach an entity-specific restriction to InnoLight or Eoptolink because the rule did not name them. MLPerf Training and Inference v6.0 repository heads were unchanged, and the closed inference submitter list still showed no matched current Chinese accelerator. InferenceX coverage still did not supply an equivalent Chinese result. No independent latest-versus-latest parity claim was added. Historical news, reports, forecasts, evidence labels and prior run records were preserved.","source_ids":["pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","amec-cseac-2026-09-05","acm","ciomp","empyrean","duv-ft","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","guowang","h3c","hwatsing"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with September 7–14 overlapping English and Chinese searches. Added H3C primary evidence for its reported Xiaohongshu DDC deployment, G-Link and ODCC participation, UXM module support and open optical-interconnect direction. The disclosure did not provide a matched InfiniBand workload benchmark, cluster topology, switch-silicon provenance or independent customer data. Added reporting sourced to Hwatsing's company announcement that its Master-P510APEX panel-level CMP system was shipped to an advanced-packaging production line; kept shipment distinct from customer acceptance, yield, uptime, throughput and volume evidence. Reopened Guowang's patent record and the current H3C and Hwatsing company pages; Guowang gained no new production evidence. Search-only coverage of the other priority profiles did not advance their review dates. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current public coverage. The MLPerf repositories had no newer result push and no current Chinese accelerator submitter was found; InferenceX listed current GB300, MI355X and TPU v7 systems but no equivalent current Chinese hardware result. No matched latest-versus-latest parity claim was added. Historical reports, corrections, forecasts, evidence labels and prior run records were preserved.","source_ids":["h3c-odx-2026-09-04","hwatsing-panel-cmp-2026-09-09","guowang","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-14","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","huawei","hwatsing","iluvatar","innolight","metax","naura","rslaser","smic","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and retained the September 11 period-to-date edition unchanged. Audited the ten selected in-period events and ran overlapping English and Chinese gap searches across accelerators, software, memory, optical interconnects, lithography and packaging equipment. Added Biren's September 7 vendor account of an August 28 ecosystem meetup as a canonical source and event: it says TorchSUPA uses PyTorch PrivateUse1 and supports common APIs, collectives and FSDP2, graph capture and profiling. No public code repository was identified, so the record does not infer operator coverage, maintainability, upstream acceptance, performance or CUDA parity. The complete briefing preserves all historical reports and states that no dated forecast moved. It uses the current September 14 benchmark review: no matched contemporary Chinese result against GB300, MI355X or TPU v7 was verified in MLPerf v6.0 or InferenceX coverage. Because systematic coverage began September 11, the September 7-13 reconstruction remains incomplete for private deployments, unindexed disclosures, tenders and individual benchmark artifacts. Search-only coverage did not advance other company review dates.","source_ids":["biren-torchsupa-2026-09-07","biren-2026-h1","biren-glm-2026-09-07","biren-shanghai-lab-2026-09-10","pytorch-cambricon-2026-09-07","pytorch-china-2026-09-09","hbm-reuters","cioe-optics-2026-09-11","enflame-ipo","hwatsing-panel-cmp-2026-09-09","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","iluvatar","jcet","kingsemi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 8-15. No new qualifying in-window accelerator, memory, optical-interconnect or semiconductor-equipment event was verified. A filing-gap review found three material August disclosures that were missing from the canonical record. AMEC's company report says three ICP etch families were in volume production on lines at more than 50 customers, with stable production use or repeat orders for additional advanced etch systems and repeat production orders after onsite validation of tungsten deposition systems at key memory and logic customers. NAURA reports more than 1,000 cumulative deliveries each for etch, PVD and vertical furnaces, RMB 20.161 billion of first-half revenue, and rollout of hybrid-bonding and high-end TSV electroplating tools. JCET reports XDFOI volume production, customer validation of silicon-photonics engine samples, and 40.4% year-over-year growth in computing-electronics revenue. Each claim remains vendor evidence: unnamed customers, unspecified process nodes, and undisclosed yield, capacity, uptime, throughput or production volume limit the conclusions. Reviewed Iluvatar and Kingsemi's current official pages without accepting a new production or benchmark claim. Rechecked MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was found. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the August events are explicitly late-added rather than dated as September events.","source_ids":["amec-2026-h1-report","naura-2026-h1-report","jcet-2026-h1-report","iluvatar","kingsemi","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","kunlunxin","metax","moore-threads"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 9-16. Added Moore Threads' September 9 primary announcement that JD Cloud plans a 100,000-card full-function GPU cluster. The record treats this as a plan rather than delivery evidence because no schedule, GPU model, completed card count, topology, interconnect, power, recovery or workload result was disclosed. Backfilled Moore Threads' August half-year report: company-reported revenue was RMB 1.736 billion, attributable net loss was RMB 11.56 million, and a RMB 660 million KUAE sales contract was fully delivered and recognized. Its Dense and MoE MFU, scaling-efficiency and effective- training-time figures remain vendor measurements without a matched workload, power or current foreign-hardware comparison. Added Kunlunxin's downloadable FlagOS MiniCPM5-2B deployment image and MetaX's ATRIA Dawn Preview Day-0 adaptation plus reported PyTorch operator coverage as compatibility evidence, not performance or CUDA parity. Rechecked MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates. Historical reports, corrections, forecasts and evidence labels were preserved; the half-year filing is explicitly late-added.","source_ids":["moore-threads-2026-h1-report","moore-threads-jd-cloud-2026-09-09","kunlunxin-minicpm-2026-09-10","metax-atria-2026-09-14","mthreads","kunlunxin","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","skyverse"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 10-17. Added Reuters' September 17 account of Huawei's new launch targets for the 960DT in Q1 2027 and Ascend 960PR in Q3 2027, plus Huawei's claim that it had shipped more than 1,000 supernodes to more than 370 customers. The record preserves the older dated roadmap and treats the new dates as expected milestones, not delivery evidence. Huawei did not identify customers, state the number of accelerators per shipped system or disclose acceptance, utilization, workload, power, reliability or interconnect-performance results. Added Huawei's September 10 account of China Unicom deploying Ascend compute and a 384-NPU supernode and adapting Ascend 950 for Unicom Cloud; this is named-deployment evidence but remains a vendor account without independent operating metrics. Reviewed Skyverse's current official portfolio and news pages without accepting prototype or production claims from image-only reporting. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["huawei-connect-2026-09-17","huawei-china-unicom-2026-09-10","skyverse","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-18","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","smee"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 11-18. Added Reuters' September 18 report that CXMT plans a NAND R&D production line at its new Beijing plant and has discussed prospective AI-storage customers. The event is recorded as a research and trial- production plan, not commercial output: Reuters could not establish an operating date or a commitment to volume manufacturing, and CXMT did not comment. Re-reviewed Reuters' materially expanded September 17 Huawei report and added a companion event for the company's statements enough AI computing equipment to satisfy domestic demand, Ascend 950 systems have entered commercial use, an Ascend 960 supernode is planned for up to 4,096 processors and Peerium is intended eventually to connect as many as one million. These are reported company claims; Huawei did not disclose capacity, orders, customer acceptance, utilization, workload, power, reliability or interconnect-performance data. Reviewed SMEE's current official product page; no new qualified production milestone was accepted. RSLaser and SIOM searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-nand-2026-09-18","huawei-connect-2026-09-17","smee","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-19","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","thead"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 12-19. Added an independent teardown of one Eoptolink 400G DR4+2km module that directly identified a Marvell IN010C50 DSP and several Analog Devices power and analog parts. The record excludes the author's inferred laser identification and limits the finding to this anonymous, single sample with no disclosed manufacturing date; it does not establish the bill of materials for current 800G or 1.6T production. The oldest-profile review reopened T-Head's current official page and the existing May Reuters report, then backfilled the materially under-recorded company claims of more than 560,000 Zhenwu units shipped, deployments at more than 400 external customers across 20 industries, the 128-accelerator Panjiu AL128 rack and expected V900 and J900 roadmap dates. Product mix, acceptance, utilization, workload performance, power, fabrication and memory provenance remain undisclosed. RSLaser and SIOM searches again produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads, the closed inference submitter directory and InferenceX's current public coverage. No matched contemporary Chinese result against current GB300, MI355X or TPU v7-class systems was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["eoptolink-400g-teardown-2026-09-19","m890","thead","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-20","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 13-20. Added Reuters' September 20 account of CXMT's statement that its fifth-generation DRAM platform entered mass production using quadruple patterning and an 11.95 nm active-area half-pitch, and that two 24 Gb LPDDR5X products made on the platform are in mass production. The record treats these as company claims and preserves Reuters' distinction between gross dies per wafer and the proportion passing final tests: the claimed at-least-50% improvement over the fourth- generation platform does not disclose yield, reliability, customer qualification or output volume. CXMT said Chinese equipment makers contributed to critical process steps but did not name them or establish a fully domestic process flow. The event is a commodity-DRAM milestone, not evidence of HBM volume or performance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage, which now names Rubin alongside GB300, MI355X and TPU v7. No matched contemporary Chinese accelerator result was verified. Search-only priority coverage did not advance other company review dates; historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["cxmt-g5-dram-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 14-21. Late-added a July 22 Pengcheng Laboratory preprint that measures Ascend 910A/B/C on scientific kernels and five application studies. The record captures a favorable matched quantum-simulation case: a tuned 910C implementation completed a 30-qubit, 30-layer circuit in 11.4 seconds versus 14.3 seconds on an A800 using cuQuantum. It also preserves the paper's less favorable evidence: vector- and memory-bound kernels remained constrained, several comparisons used different algorithms or implementations, and workload-specific optimization was central. The study uses export-limited A800/H800 controls rather than current frontier hardware and does not test AI model training or inference, so it does not change the latest-versus-latest assessment. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked unchanged MLPerf Training and Inference v6.0 repository heads and InferenceX's current public coverage; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or the newly listed Jalapeno systems was verified. Search-only priority coverage did not advance other company review dates. Historical reports, corrections, forecasts and evidence labels were preserved.","source_ids":["ascend-science-2026-07-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-21","status":"updated","companies":["cxmt","eoptolink","huawei","metax","ymtc"],"summary":"Ran the weekly helper for the previous fully closed Monday-Sunday period and created the complete September 14-20 edition without replacing the earlier September 7 period-to-date or complete reports. Audited all six selected in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, memory, optical interconnects, software and manufacturing equipment. The synthesis treats Huawei's accelerated 2027 roadmap, reported system scale and Peerium targets as company evidence rather than delivered workload performance, and makes Huawei's inability to meet Chinese demand the clearest capacity constraint. It distinguishes CXMT's fifth-generation commodity-DRAM and LPDDR5X mass-production claims from tested yield and HBM progress, and its planned NAND R&D line from commercial output. It uses the single Eoptolink 400G teardown only as sample-specific component evidence. The pre-period Ascend scientific- computing paper is background because it was public by the period end; it does not enter news_ids or establish model-training or inference parity. Rechecked unchanged MLPerf v6.0 repository heads and current InferenceX coverage. No matched current Chinese result against Rubin, GB300, MI355X or TPU v7 was verified, and no dated forecast changed. Coverage remains incomplete for private and unindexed evidence, tenders, individual benchmark artifacts and undisclosed production or customer metrics.","source_ids":["ascend-science-2026-07-22","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","cxmt-nand-2026-09-18","eoptolink-400g-teardown-2026-09-19","huawei-connect-2026-09-17","inferencex","metax-atria-2026-09-14","mlperf-inference","mlperf-training"]},{"date":"2026-09-22","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","thead","ligent"],"summary":"Screened all priority-1 profiles and the next three oldest-reviewed non-priority profiles with overlapping English and Chinese searches covering September 15-22. Added Reuters' September 22 report that Alibaba unveiled T-Head's Zhenwu V900 and now targets Q1 2027 mass production and commercial release, earlier than the Q3 2027 roadmap reported in May. Preserved that older dated roadmap and recorded the new three-times-M890 performance and 500,000-chip cluster claims as reported targets rather than workload evidence: precision, model, power, topology, recovery, fabrication and memory data were not disclosed. Added Ligent Technologies as a tracked photonics supplier after its HKD 5.67 billion Hong Kong IPO, with planned allocations of about 52.9% of net proceeds to faster optical-product and chip R&D and 25.1% to capacity. The record separates capital raised and a vendor-listed 400G/800G/1.6T portfolio from verified high-speed shipment mix, yield, customer qualification and domestic component provenance. RSLaser, SIOM and Zhiqi searches produced no retrievable new primary evidence, so their review dates did not advance. Rechecked the unchanged MLPerf Training and Inference v6.0 repository heads and submitter directories plus InferenceX's current dashboard; no matched contemporary Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified. Search-only priority coverage did not advance other company review dates.","source_ids":["alibaba-v900-2026-09-22","ligent","ligent-ipo-2026-09-22","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-23","status":"partial","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened these 15 companies with a September 16-23 overlap, not a full source revalidation of every profile. Backfilled Huawei's primary deployment/testing disclosure and interconnect design; retained vendor labels and historical roadmaps. Only Huawei's profile review date advanced. English and Chinese searches for the other priority companies largely returned previously recorded, older or irrelevant results. RSLaser and Zhiqi searches and the FT search excerpt did not establish new qualification, shipment or production evidence; the full FT article could not be retrieved. SIOM's official about page returned 502 and targeted searches did not resolve the gap. These oldest profiles remain pending substantive review. Rechecked MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and top-level submitters, Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 and closed submitters, plus InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing was changed.","source_ids":["huawei-peerium-2026-09-17","duv-ft","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-24","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi","robotechnik"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 17-24. Search results for those 15 profiles were previously recorded, older, immaterial or insufficient to establish a new qualification, delivery, production or benchmark milestone, so their profile dates did not advance. Reviewed RoboTechnik's full September 21 Hong Kong prospectus and added the company as a priority-2 silicon-photonics equipment supplier. The company reports RMB 2.0 billion of SiPh equipment backlog at June 30 and says mainland production rose from 28.1% of units in 2025 to 62.5% in the first four months of 2026. The record qualifies these as company evidence and notes that core ficonTEC operations, R&D and know-how remain geographically distributed. The proposed offering could provide up to an estimated HKD 4.961 billion net at the maximum price, with 20% planned for R&D and 40% for capacity; final pricing, proceeds and execution were pending when reviewed. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1, their current submitter directories and InferenceX's published coverage. No matched current Chinese accelerator comparison was verified; individual result artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["robotechnik-prospectus-2026-09-21","robotechnik-offering-2026-09-20","mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-25","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 18-25. Results were previously recorded, older, immaterial, inaccessible or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1 plus their current submitter directories. InferenceX has added newer model coverage, but no comparable Chinese accelerator result was identified in its published hardware coverage. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-26","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 19-26. Results were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production volume or yield, tool acceptance, customer adoption, capacity-changing financing or software/interconnect milestone. SIOM's September 22 fused-silica study was read in full; it addresses radiation-driven degradation after ion-beam figuring for space and high-power-laser optics, not a demonstrated lithography-tool delivery, production qualification or AI-hardware milestone. Company profile review dates therefore did not advance. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.0 head 4d3916ac9cf474b679cdfcf492d43a0559418ad1. InferenceX's published hardware coverage now names Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño but still supplied no comparable current Chinese accelerator result. Individual result artifacts were not exhaustively reviewed. No dated forecast, historical news event or briefing changed.","source_ids":["mlperf-training","mlperf-inference","inferencex"]},{"date":"2026-09-27","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 20-27. Backfilled Huawei's September 17 primary disclosure that its Hi-ONE NPO engine has 7.2T transmission capacity per engine and company-claimed mass-production capability. The record does not infer a completed Ascend 960 system, shipped volume, customer acceptance, production yield, component provenance, reliability or workload performance. Only Huawei's company review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Updated the current standardized benchmark review from MLPerf Inference v6.0 to the September 16 v6.1 results. Its 30-participant submitter list contains no current Chinese accelerator vendor, so it still does not supply a matched latest Chinese-versus-foreign comparison; standardized submissions are submitter-run rather than automatically independent. MLPerf Training v6.0 remains the latest published training round while v6.1 submissions are due in October. Rechecked InferenceX's current coverage and found no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-hi-one-2026-09-17","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 21-28. Added NAURA's September 24 statement that its atomic-layer-etch equipment has achieved customer process use and linked the company's undated product disclosure that the tool entered a well-known customer's production line and completed process validation. Recorded this as vendor evidence, not proof of advanced-node volume production, named-customer acceptance, installed units, yield, uptime, throughput or domestic component provenance. Also recorded NAURA's warning that material-delivery tightness at individual suppliers may pressure some order deliveries over the next three to six months; affected components, suppliers and order value were not disclosed. Only NAURA's profile review date advanced. Results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus their current submitter directories; no matched latest Chinese-versus-foreign accelerator comparison was verified. InferenceX still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["naura-ale-primary","naura-ale-2026-09-24","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-28","status":"updated","companies":["ligent","naura","robotechnik","thead"],"summary":"Ran the weekly briefing helper for the closed September 21-27 period, retained all earlier editions and audited the four selected canonical events against their original sources. Supplemented the corpus with Alibaba's September 22 primary announcement, which supports the V900 and M890 Supernode roadmap while also recording that medium- and long-term demand exceeds supply capability and wider data-center supply-chain shortages constrain expansion. The source did not disclose delivered configurations, customer acceptance, utilization, reliability, power or workload results. Bounded English and Chinese gap searches across accelerators, optical interconnects and capital, memory, networking and manufacturing equipment did not establish an additional qualifying in-period event. A clearly labeled September 28 post-period check of MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX found no matched current Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed disclosures, tenders, customer acceptance, volume, yield, utilization, power and component provenance.","source_ids":["alibaba-v900-2026-09-22","alibaba-v900-primary-2026-09-22","ligent-ipo-2026-09-22","ligent","naura-ale-2026-09-24","naura-ale-primary","robotechnik-offering-2026-09-20","robotechnik-prospectus-2026-09-21","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-29","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 22-29. Added the September 25 bipartisan Senate proposal that would prohibit federal procurement of covered InnoLight and Eoptolink optical transceivers for national-security systems, including covered modules using their firmware, software or components. The bill provides a five-year transition and limited renewable waivers. Recorded it as proposed legislation, not an enacted, immediate or market-wide ban. Reuters reported industry concern about data-center supply disruption and said neither company commented for its report. Only Eoptolink and InnoLight received new profile review dates; results for the other 13 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. Rechecked unchanged MLPerf Training v6.0 head eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 head 10ecdffda3bb94d71f0203a6ca8e20c17943f27c plus InferenceX's published coverage. No matched latest Chinese-versus-foreign accelerator comparison was verified; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["us-optical-transceiver-bill-primary-2026-09-25","us-optical-transceiver-bill-reuters-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-09-30","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 23-30. Added DeepSeek contributors' September 30 open-source TileLang backend for Ascend 950, which includes native code generation, scheduling, synchronization, SIMT/SIMD vector programming and kernel examples. Reuters reports that DeepSeek and Huawei also worked on compute and communication libraries and a 128-chip Ascend 950 supernode solution. Recorded this as inspectable software and integration evidence, not proof of broad operator coverage, production stability, adoption, power efficiency, CUDA equivalence or matched workload performance. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421. MLPerf Inference v6.1 advanced to 4bb63cd28eb136be221bc801a9e38c66c8a7553f with previously missing submission documentation; its 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["tilelang-ascend-950-2026-09-30","deepseek-huawei-tilelang-reuters-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 24-October 1. Added Huawei's October 1 statement that China's advanced-semiconductor capacity remains very limited, Ascend AI chips draw on the same capacity used for smartphone processors, and domestic advanced-chip manufacturing still relies on DUV while manufacturers work toward EUV. Recorded this as direct fabrication-constraint evidence without inferring capacity, yield, node, equipment provenance or SMIC production confirmation. Backfilled Huawei's September 18 primary announcement of an Ascend 950 cloud cluster service, including vendor claims of 40 days of stable training, recovery within 10 minutes and 20% higher token throughput than its prior generation. The announced September 30 Chinese commercial date was not converted into verified availability because no separate public confirmation, configuration, workload, quality target or customer evidence was found. Only Huawei's profile review date advanced; results for the other 14 profiles were previously recorded, older, immaterial or insufficient to establish a new qualified delivery, production, acceptance, financing, software or interconnect milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; the latter's 30 closed submitter directories still include no current Chinese accelerator vendor. InferenceX's published hardware coverage still supplied no comparable current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","huawei-cloud-ascend-950-service-2026-09-18","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","iluvatar","innolight","kunlunxin","ligent","metax","moore-threads","naura","robotechnik","rslaser","thead","ymtc","yuliangsheng","zhiqi"],"summary":"Ran the monthly briefing helper for the fully closed September 1-30 period and retained the September 11 period-to-date edition unchanged. Audited all 30 canonical in-period events and their original sources, then ran bounded English and Chinese gap searches across accelerators, software, memory, networking, optical interconnects, lithography, manufacturing equipment and capacity-changing capital. No additional qualifying September event was established. Because systematic tracking began September 11, the September 1-10 period remains a reconstruction rather than a complete contemporaneous census. The complete edition synthesizes stronger evidence of accelerator commercialization, named deployments, system roadmaps, upstream and inspectable software work, optical-module scale and intermediate equipment qualification. It keeps HBM scarcity, constrained accelerator supply, foreign optical components, incomplete tool metrics and absent matched current benchmarks as explicit dependencies. The October 1 benchmark review found no comparable current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeno in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical report changed, and company profile review dates did not advance for this synthesis.","source_ids":["alibaba-v900-2026-09-22","amec-cseac-2026-09-05","biren-2026-h1","cioe-optics-2026-09-11","cxmt-g5-dram-2026-09-20","deepseek-huawei-tilelang-reuters-2026-09-30","enflame-ipo","hbm-reuters","huawei-connect-2026-09-17","naura-ale-2026-09-24","us-optical-transceiver-bill-primary-2026-09-25","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-01","status":"updated","companies":["aishengna","amec","biren","cambricon","cxmt","enflame","eoptolink","h3c","huawei","hwatsing","innolight","jcet","ligent","metax","moore-threads","naura","robotechnik","thead","ymtc","yuliangsheng"],"summary":"Ran the quarterly briefing helper for the fully closed July 1-September 30 period and retained the September 11 period-to-date edition unchanged. Audited the 37 previously canonical in-period events and their original sources, then ran bounded overlapping English and Chinese gap searches across accelerators, software, lithography, foundries, process equipment, memory, packaging, networking, optics and capital. Added three missing Q3 events: planned CXMT and YMTC capacity expansion paired with reported ASML DUV reliance; disputed evaluations of AMEC etch tools at Samsung and SK Hynix China fabs, with both memory companies' denials and the absence of a deployment decision retained; and Innolight's completed Hong Kong offering with HKD 52.891 billion of estimated net proceeds. Added a Reuters source confirming the existing small-volume CXMT HBM3E report. Only AMEC, CXMT, Innolight and YMTC received new profile review dates; briefing synthesis did not advance other company dates. Because systematic tracking began September 11, July, August and September 1-10 remain a reconstruction rather than a complete contemporaneous census. The quarter strengthened evidence for Chinese-controlled designs, software, aggregate systems, optical modules and selected equipment while leaving replenishable domestic HBM, lithography, foreign optical components and upstream inputs unresolved. The October 1 benchmark review found no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño in MLPerf Training v6.0, MLPerf Inference v6.1 or InferenceX's published coverage; individual artifacts were not exhaustively audited. No dated forecast or historical briefing changed.","source_ids":["alibaba-v900-2026-09-22","amec-2026-h1-report","amec-evaluation-reuters-2026-08-05","ascend-field","ascend-science-2026-07-22","china-memory-expansion-reuters-2026-07-24","cioe-optics-2026-09-11","cxmt-hbm-reuters-2026-08-31","duv-reuters","hbm-reuters","huawei-connect-2026-09-17","innolight-hk-listing-2026-07-29","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-02","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","siom","zhiqi"],"summary":"Screened the 12 priority-1 profiles and the three oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 25-October 2. Backfilled CXMT's missing August 29 LPDDR6 event using Reuters and the company's September 7 detailed disclosure: CXMT and Xiaomi said the memory entered mass production for the Xiaomi 18 Fold, while CXMT reported a 16 Gb die, 16 GB package and up to 12,800 Mbps. The record preserves the company's internal-test attribution and does not infer volume, yield, reliability at scale, fabrication- equipment provenance or HBM capability. Added Reuters' October 1 account of a reported five-year, roughly $7 billion Tencent lease for about 100,000 advanced AI chips at Oracle data centers in Southeast Asia. It is recorded as system-level foreign-compute dependency evidence, not a verified delivery: Reuters could not independently confirm the Financial Times report, and chip models, schedule, utilization and workloads were undisclosed. Reviewed SIOM's official late-September research index and found research advances but no commercial lithography-source qualification. Reviewed Fujing Technology's September 2 primary disclosure, added Zhiqi's Chinese legal name and its reported first-half growth while retaining unknown revenue, customer, qualification, precision, yield and production data. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Results for the other priority profiles were previously recorded, immaterial or insufficient for a new qualified milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f; InferenceX still supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["cxmt-lpddr6-primary-2026-09-07","cxmt-lpddr6-reuters-2026-08-29","tencent-oracle-ai-chip-lease-reuters-2026-10-01","zhiqi-fujing-ir-2026-09-02","siom","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-03","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","acm","ciomp","empyrean"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 26-October 3. Added ACM Research Shanghai's October 1 primary filing: it reported RMB 17.073 billion of orders on hand at September 29, up 88.2% year over year. The record retains the filing's cautions that the unaudited aggregate combines future deliveries with delivered equipment awaiting revenue recognition, may not convert fully, and supplies no customer, product, fab, qualification, acceptance, cancellation or domestic-content split. Backfilled Empyrean's August 29 half-year filing: the company says products entered core design flows at leading domestic design and manufacturing customers and supported mass production, and reports ALPS certification at 8 nm, 5 nm and 4 nm, Patron at 14 nm, and Argus DRC/LVS at 28 nm. Customers, foundries, tape-outs, covered rule decks, sign-off correlation, production volume and evidence of a complete foreign- tool displacement remain undisclosed. Reviewed CIOMP's current official research and news indexes; no new production-lithography optics milestone was established. Current primary evidence for RSLaser remained unavailable, so its profile date did not advance. Backfilled InferenceX's June 9 analysis of DeepSeek V4 day-one inference on a 16-rank Ascend 950DT setup. The analysis verifies working optimized inference and profiler evidence, but its public article does not give a same- harness numerical comparison with H200 or B200 and says that comparison was future work; the profiler images are attributed to SemiAnalysis and Huawei. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. No matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño was verified, and individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","empyrean-2026-h1-report","ciomp","inferencex-ascend-950dt-deepseek-v4-2026-06-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-04","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","h3c"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 27-October 4. Added H3C's September 29 UniPoD S81000 X1 announcement after reviewing the company's current news index. H3C says the rack-scale system supports 40 unnamed domestic accelerators in one scale-up domain, using four dedicated switch trays in a single-stage CLOS topology to provide 448 GB/s peer-to-peer bandwidth between any two cards, 5,760 GB of unified-address memory and up to 28 PFLOP/s at FP8. The record keeps these as vendor specifications rather than a benchmark: H3C did not identify the accelerator or switch silicon, distinguish dense from sparse FP8, disclose power, latency, utilization, customer acceptance, shipment volume or workload results, or publish a matched comparison with NVLink, InfiniBand or a current foreign supernode. Reviewed Biren's current official news index; its newest indexed item remained the September 10 disclosure already captured by the Observatory. Current primary evidence for RSLaser remained unavailable, and Guowang's public patent record did not establish a new production milestone. Search-only coverage of the other priority profiles did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["h3c-unipod-s81000-x1-2026-09-29","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","hwatsing"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 28-October 5. Added Hwatsing's September 30 announcement that it completed a private placement of 16,080,508 A shares at RMB 236 per share, raising RMB 3.795 billion before expenses. The company says proceeds will support a Shanghai semiconductor-equipment R&D and manufacturing base, Kunshan wafer-reclaim expansion and high-end equipment R&D. This is financing evidence, not installed-capacity or qualified-tool evidence: schedules, tool mix, domestic-component content, yield, uptime and realized output remain undisclosed. Added Huawei and Qualcomm's October 5 multi-year patent cross-license covering 5G, compute, AI and networking, plus Qualcomm's planned purchase of certain Huawei US patents subject to regulatory approval. The parties did not disclose terms, patents or specific product implementations; the agreement is not a chip-supply or manufacturing arrangement and does not prove that Ascend uses named Qualcomm technology. Reviewed Biren's current official news index, which remained unchanged from September 10. Current primary evidence for RSLaser remained unavailable, and Guowang's public record did not establish a new production milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's published coverage supplied no matched current Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["hwatsing-placement-2026-09-30","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-05","status":"updated","companies":["acm","h3c","huawei","hwatsing"],"summary":"Published the complete September 28-October 4 weekly briefing after auditing all six canonical in-period events and their original sources, plus bounded overlapping English and Chinese gap searches across accelerators, software, networking and scale-up systems, lithography and fabrication capacity, manufacturing equipment, capital and benchmarks. The edition separates inspectable Ascend software, H3C's announced 40-accelerator rack, ACM's unaudited order book and Hwatsing's completed financing from operational performance, customer qualification and realized output. It records Huawei's statement that scarce domestic advanced-chip capacity is shared between Ascend and smartphone processors and that Chinese advanced manufacturing still relies on DUV, along with the reported and unverified Tencent lease for about 100,000 foreign AI chips in Southeast Asia. No additional material in-period event was established. MLPerf Training v6.0, MLPerf Inference v6.1 and InferenceX still supplied no matched current Chinese comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño; individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["acm-orders-2026-10-01","deepseek-huawei-tilelang-reuters-2026-09-30","h3c-unipod-s81000-x1-2026-09-29","huawei-advanced-capacity-reuters-2026-10-01","hwatsing-placement-2026-09-30","tencent-oracle-ai-chip-lease-reuters-2026-10-01","tilelang-ascend-950-2026-09-30","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-06","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 29-October 6. Added The Wire China's October 4 original reporting that YMTC responded to lost access to critical U.S. chipmaking tools by buying Dutch and Japanese equipment. The same report says at least one YMTC innovation was licensed to Samsung. These are important resilience and IP signals, but not evidence of a replenishable fully domestic production flow: the report did not identify equipment suppliers or models, quantify the domestic-versus-foreign tool share, or disclose the licensed patent, terms, production use, yield or volume. Reviewed Biren's current official news index, Iluvatar's company and developer news indexes, RSLaser's available public product record and Guowang's public record; none established a new qualified production or benchmark milestone. Search-only coverage did not advance the other profile dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's latest published article remained its September 28 memory-analysis item and supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["ymtc-wire-china-2026-10-04","biren","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-07","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","jcet"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering September 30-October 7. Backfilled JCET's September 3 board-approved proposal to raise up to RMB 6.5 billion, including RMB 1.5 billion for a high-performance-computing advanced-packaging expansion, RMB 1.0 billion for high-density, high-capacity memory-chip system-level packaging and test, and additional funds for wafer-level and power-module packaging. This is a conditional financing and investment plan, not completed funding, installed capacity or operating output. State-asset and shareholder approvals, Shanghai Stock Exchange review and securities-regulator registration remained prerequisites; the filing did not disclose construction schedules, customer qualification, domestic-input provenance, yield, throughput or realized output. Current searches for the remaining profiles did not establish another new qualified production, deployment or benchmark milestone, so their review dates did not advance. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed.","source_ids":["jcet-private-placement-plan-2026-09-04","biren","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-08","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 1-8. Huawei's October 1 Mate 90 and advanced-capacity disclosure was already canonical. Biren's current official news index remained unchanged from September 10. Current public records for RSLaser, Guowang and Kingsemi did not establish a new qualified production, deployment or customer-adoption milestone. Search-only coverage did not advance any profile review date. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["huawei-advanced-capacity-reuters-2026-10-01","duv-ft","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-09","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","iluvatar"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 2-9. Backfilled reporting tied to The Wire China's October 4 profile that an unnamed YMTC employee said the company had raised chip prices, shifted production toward higher-margin products and expected the global memory shortage to last at least another three years. This is reported market and allocation evidence, not company guidance, a quantified price or capacity disclosure, or proof of added sovereign supply. The account did not identify affected products, price changes, capacity allocation, customer contracts or whether the forecast referred specifically to NAND rather than memory broadly. Only YMTC's profile date advanced. Current public records for RSLaser, Biren, Guowang and Iluvatar did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-wire-china-2026-10-04","ymtc-shortage-techradar-2026-10-06","biren","guowang","iluvatar","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-10","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","biren","guowang","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 3-10. Added Biren's October 8 primary filing that its unnamed next-generation product completed bring-up and entered post-silicon validation. The company expects customer sampling and volume production over the coming quarters, subject to satisfactory validation and qualification; this is therefore an engineering milestone rather than customer adoption, qualification or production. Also recorded Biren's best-efforts placement of 130 million H shares at HK$31.08 each, for estimated net proceeds of HK$4.0197 billion. The filing allocates 70% to supply-chain procurement, production readiness and commercialization, 20% to research, software and ecosystem work, and 10% to working capital, but listing approval remained a prerequisite and the transaction might not proceed. As of September 30 the company had used HK$982.8 million, or 14%, of its prior July placement proceeds. The filing did not name the new product or disclose specifications, fabrication, memory, yield, customers, benchmarks, suppliers, reserved capacity or commercial commitments. Only Biren's profile date advanced. Current public records for RSLaser, Guowang and Kingsemi did not establish another qualified production, deployment, customer-adoption or benchmark milestone; search-only coverage did not advance their review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no new matched Chinese accelerator comparison against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","biren","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-11","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 4-11. Added the October 9 national new-quality-productive-forces guidance as a company-neutral policy event. It calls for faster AI theory and core-technology development, stronger compute, algorithm and data supply, a national integrated computing-power network and broader AI adoption across industries and devices. The same guidance requires locally appropriate planning and warns against herd-like competition, bubbles, blind investment and duplicative local expansion. It sets policy direction but provides no quantified budget, implementation schedule, procurement award or company allocation. No fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone was established for the screened companies, so no profile review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-new-quality-productive-forces-policy-2026-10-09","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","iluvatar","kingsemi"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 5-12. Backfilled Iluvatar's September 29 official interim report: first-half revenue was RMB 945.7 million, GPGPU revenue was RMB 916.1 million and the company reported more than 380 customers. Inference-product revenue was RMB 654.2 million and training-product revenue was RMB 261.9 million. The filing does not name customers or disclose unit shipments, product mix, utilization or workload results. Total gross margin fell to 17.2%, net profit was RMB 106.3 million alongside a RMB 760.2 million fair-value gain, operating cash outflow was RMB 2.786 billion and borrowings reached RMB 2.687 billion. AI-computing-solutions revenue fell 68.1% amid supply-resource coordination and delivery-schedule adjustments; this does not by itself establish a specific chip shortage or customer cancellation. Only Iluvatar's profile date advanced. Current public records for the other screened companies did not establish another fresh qualifying production, deployment, customer-adoption, financing or benchmark milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and MLPerf Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX supplied no matched Chinese accelerator comparison against current Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño systems. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["iluvatar-2026-h1-report","guowang","kingsemi","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-12","status":"updated","companies":["biren","huawei"],"summary":"Ran the weekly briefing helper for the fully closed October 5-11 period and retained all earlier editions. Audited all four canonical in-period events and their original sources, using the Observatory's daily overlapping English and Chinese searches across 16 priority and oldest-reviewed companies to check gaps. The complete edition separates Biren's reported post-silicon bring-up from validation, customer sampling, qualification and volume production; it also separates Biren's proposed HK$4.0197 billion placement from completed funding, reserved capacity and realized output. The national guidance combines stronger AI compute, algorithm, data and adoption support with controls against bubbles, blind investment and duplicative local expansion, but provides no budget, schedule, procurement award or company allocation. Huawei and Qualcomm's cross-license is recorded as global IP exchange rather than a chip-supply agreement or proof of a named Ascend dependency. MLPerf Training v6.0 and Inference v6.1 heads remained unchanged, and InferenceX supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively audited. No additional material in-period event was established and no dated forecast changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["biren-placement-2026-10-08","china-new-quality-productive-forces-policy-2026-10-09","huawei-qualcomm-patent-license-2026-10-05","huawei-qualcomm-patent-license-reuters-2026-10-05","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-13","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","kingsemi","kunlunxin","thead"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 6-13. T-Head was also updated because the accepted market source directly covered it. Reuters reported IDC first-half data showing Huawei shipped 572,000 AI accelerators in China for 28% share, with T-Head at 7%, Cambricon at 5% and Kunlunxin at 4%. Total shipments were 2.008 million, and non-GPU units rose nearly 80% to 1.016 million. The figures strengthen adoption evidence but do not identify product mix, buyers, accepted deliveries, utilization, workloads, prices or component provenance, and export restrictions make market share unsuitable as a frontier-performance comparison. Kunlunxin's October 10 primary release and RLinf's release-v0.4 repository added inspectable P800 backend evidence across single- and multi-node modes, Ray scheduling, XCCL collectives and Pi 0.5 LIBERO reproduction. Vendor task and synchronization results lack a matched foreign-hardware control, disclosed power or cost, independent reproduction or production-customer evidence. Kingsemi management reported material order growth, production pressure, capacity expansion and a shift toward front-end track and cleaning equipment; no order, output, accepted-tool, fab-node or qualification counts were disclosed, and a funded expansion project remains delayed. Only Huawei, Cambricon, Kunlunxin, T-Head and Kingsemi received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["china-ai-accelerator-share-reuters-2026-10-13","kunlunxin-rlinf-2026-10-10","rlinf-kunlunxin-support","kingsemi-capacity-2026-10-10","duv-ft","guowang","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-14","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","metax","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 7-14. Backfilled CXMT's September 28 board-approved proposal to use RMB 13.0 billion of over-raised IPO funds toward a RMB 24.1 billion technology-research project and RMB 5.0 billion toward a RMB 10.8 billion second-phase DRAM wafer back-end testing base. The proposal remained subject to shareholder approval and does not establish construction, installed test capacity, qualified output, yield or HBM volume. AMEC's October 13 filing added a proposed RMB 35 million investment in Qinghe Jingyuan within a RMB 250 million round intended partly for hybrid-bonding and atomic-beam equipment research and customer delivery. The agreement remained unsigned, and reported customer validation does not establish accepted production tools, throughput, reliability or domestic component provenance. Only CXMT and AMEC received new profile-review dates. No other fresh qualifying production, deployment, customer-adoption or financing milestone was established. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-overfund-projects-2026-09-29","amec-qinghe-jingyuan-investment-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-15","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 8-15. Backfilled Counterpoint's independent second-quarter NAND shipment estimate: YMTC held 14% by volume, ranking third behind Samsung and SK hynix, with shipments up 22% year over year and 5% quarter over quarter. Counterpoint ranked YMTC fifth by revenue because its mix remained concentrated in lower-priced consumer applications with a low share of data-center enterprise SSDs, so the volume ranking does not establish enterprise leadership, production yield, customer qualification or supply- chain sovereignty. Also added the Shanghai Stock Exchange's August 21 record accepting YMTC's application for a proposed RMB 33 billion STAR Market raise. The project was under inquiry; the record does not establish completed financing, final proceeds, installed capacity, qualified output or a fully domestic equipment and materials flow. Only YMTC received a new profile-review date. Current official or original-source checks for the other screened companies did not establish another fresh qualifying production, deployment, customer- adoption or financing milestone. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["ymtc-star-market-ipo-2026-08-21","counterpoint-ymtc-q2-shipments-2026-08-12","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-16","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","metax","moore-threads"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 9-16. MetaX's October 16 filing forecast January-September revenue of RMB 2.541-3.106 billion and attributable net profit of RMB 586.8-717.2 million, compared with a RMB 345.5 million loss a year earlier. The forecast was unaudited, and the filing said increased investment in supply-chain companies also contributed to the profit turnaround; customer names, units, workloads, utilization, fabrication and memory provenance were not disclosed. NAURA's subsidiary signed an agreement to acquire hybrid-bonding equipment developer Nuohe Jianwei for RMB 404.6 million. Closing conditions remained, and reported product validation, shipments and repeat orders do not establish named-customer acceptance, production use, throughput, yield, uptime or domestic component provenance. Counterpoint reported that YMTC was conducting R&D and testing of a second-generation 16 Gb LPDDR5 device at roughly 5,000 wafers per month with no mass-production plan. Its article also described a 16.5% current NAND shipment share without publicly reconciling that basis with the earlier 14% second-quarter estimate, so both were retained with the discrepancy disclosed. Only MetaX, NAURA and YMTC received new profile-review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard still supplied no matched Chinese result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["metax-nine-month-forecast-2026-10-16","naura-nuohe-jianwei-acquisition-2026-10-16","counterpoint-ymtc-ipo-dram-2026-10-13","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-17","status":"updated","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","moore-threads","skyverse"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 10-17. CXMT's October 17 filing forecast January-September revenue of RMB 265.0-270.0 billion and attributable net profit of RMB 131.0-132.5 billion, compared with RMB 32.084 billion of revenue and a RMB 5.280 billion loss a year earlier. The figures were preliminary and unaudited. CXMT attributed the change to higher DRAM prices, high utilization, higher output per wafer, scale and product-mix improvements, but did not disclose units, product-level revenue, yield, customer mix, HBM volume or domestic equipment and materials provenance. Backfilled the Wuxi Huishan center's reported operating milestone: all 2,000 MTT S5000 cards in phase one were operating, aggregate demand was near full load and more than 20 customers were testing or using the cluster. This is named deployment and adoption evidence, not a workload-level utilization, reliability, power or matched foreign-hardware benchmark; the claimed full-stack domestic status also lacks public component-level provenance. Only CXMT and Moore Threads received new profile- review dates. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX's current dashboard listed Rubin, GB300, MI355X, TPU v7 and OpenAI Jalapeño systems but no matched current Chinese accelerator result. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["cxmt-nine-month-forecast-2026-10-17","moore-threads-wuxi-center-2026-09-02","moore-threads-wuxi-center-primary-2026-10-16","mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-18","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 11-18. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. Secondary coverage of AMEC's three-to-five-year equipment-development plan did not establish completed tools, customer qualification or production use beyond the existing record. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]},{"date":"2026-10-19","status":"no-material-change","companies":["aishengna","amec","cambricon","cxmt","enflame","eoptolink","huawei","innolight","naura","smic","ymtc","yuliangsheng","rslaser","guowang","skyverse","smee"],"summary":"Screened all 12 priority-1 profiles and the four oldest-reviewed remaining profiles with overlapping English and Chinese searches covering October 12-19. The sweep found no newly verified source establishing a material change in qualified deliveries, production volume or yield, named-customer acceptance, HBM provenance, tool deployment, or current accelerator performance. CXMT's chairman proposed a RMB 1.0-2.0 billion share buyback for employee incentives or capital reduction, but the proposal remained subject to board action and did not finance or establish production capacity, qualified output or HBM volume. Proposed Chinese technology-export controls were not attributed to a tracked company's operating change. No company profile-review date advanced. MLPerf Training v6.0 remained at eabf23a07b2a0c60a289ff871dc3a46fff0d0421 and Inference v6.1 remained at 4bb63cd28eb136be221bc801a9e38c66c8a7553f. InferenceX still supplied no matched current Chinese accelerator result against Rubin, GB300, MI355X, TPU v7 or OpenAI Jalapeño. Individual benchmark artifacts were not exhaustively reviewed. No dated forecast or historical briefing changed. Coverage remains incomplete for private deployments, unindexed Chinese disclosures, tenders, customer acceptance, production volume, yield, utilization, reliability, power and component provenance.","source_ids":["mlperf-training","mlperf-inference-6-1","inferencex"]}]